The African Investment Firm
After Headcount.

We reconstructed the organisational architecture of 38 African investment firms, from boutique seed funds to control private equity houses, identified the minimum viable structure of the fund as an institution, and then examined what happens to that structure when workflow becomes abundant. The result is not a forecast. It is a description of firms that already exist.

The table below is the whole argument in dashboard form: the role families of African private capital, the status of each under AI subsumption, and the seat that did not exist before. The status markers used here recur throughout the dispatch: green means the seat remains and the defining act stays human; amber means the seat or its workflow is transformed, with the machine producing and a human owning; red means the dedicated seat is subsumed, with the function persisting as infrastructure; blue marks a newly created seat.

A paradox should be noted before anything else is said. This dispatch was itself produced by the methodology that it describes. The organisational reconstruction presented in the pages that follow was not built through insider access, consulting engagements, or years of industry immersion. It was built by pointing AI systems at the material that the 38 firms in our sample publish about themselves, which is to say their About pages, their team pages, their professional profiles, and their deal announcements, and by reading that material at machine speed as organisational telemetry rather than as branding. The work that would once have required an analyst team and a research budget measured in months was performed by a small operator and an AI stack in a matter of days. In other words, the elephant mapping that Dispatch 01 of this series described as the first stage of subsumption has now been executed, for real, on an entire industry, and the reader is holding the output. Every firm in this sample emitted the visibility exhaust from which its own institutional geometry was reconstructed. This dispatch is therefore not only an analysis of institutional compression; it is a demonstration of institutional compression, and the reader should weigh its findings in the knowledge that the method and the thesis are the same thing.

The opening dispatch of this series, AI Subsumption and the New-Age Disruptor, defined the phenomenon that this whole body of work examines. Subsumption means institutional replacement through compression, and it is not the same thing as disruption. Disruption is competition inside an existing market structure, in which a new entrant competes better, prices lower, or delivers faster while still obeying the same institutional logic as the incumbent. Subsumption occurs when the underlying economic and organisational assumptions of the legacy model itself begin to collapse underneath a new operating architecture. The thesis of that opening dispatch can be stated in a single line: under industrial logic, scale preceded capability, whereas under AI-native logic, capability can now precede scale.

Dispatch 01 opened with a thought experiment. An AI-native operator wakes up on a Monday morning, decides that private equity appears structurally inefficient, and begins to study the industry's architecture, which is to say its analyst pyramids, its reporting cycles, its diligence workflows, and its committee structures, asking one question of every layer it examines: which parts of this architecture exist only because older systems required humans to move information between nodes?

This dispatch performs that study for real. It takes the operator's Monday-morning question and executes it empirically, on the fund industry itself, in the markets in which Odit Frontier Partners works. In the vocabulary of Dispatch 01, what follows is an elephant mapping: a systematic reconstruction of what African investment funds actually are, layer by layer, undertaken so that the subsumption zones, meaning the layers that exist only because information once had to be moved by hand, can be identified from evidence rather than from intuition. The questions this dispatch answers are therefore the series' own questions, applied to one specific institution. What, in the actual architecture of the African fund, is subsumable? What is not subsumable, and why not? And what does the fund look like after subsumption has run its course?

Mapping the elephant meant reconstructing it. The sample comprises 38 investment firms that are active across more than a dozen African markets, and it spans boutique seed funds, institutional early-stage venture firms, pan-African growth investors, impact vehicles anchored by development finance institutions, control private equity houses, family and corporate capital vehicles, and programme platforms. Fund sizes in the sample range from below twenty million dollars to several hundred million dollars, and team sizes range from four people to more than forty.

For each firm, we reconstructed the organisation not from its titles but from its decision rights. We asked who originates transactions, who sits on the investment committee, who leads deals, who holds board seats at portfolio companies, who owns the relationships with limited partners, who signs the financial statements, who is accountable for compliance, and who merely supports the people who do these things. The sources were public ones: team pages, professional biographies, deal announcements, promotion histories, regulatory disclosures, appraisal documents published by development finance institutions, and the firms' own published descriptions of how they work. Every reconstructed structure carries a confidence grading in the underlying research, and any reporting relationship that could not be confirmed is treated throughout as an inference rather than as a fact.

Three disciplines govern what appears in this dispatch.

• No firm is identified except where it has publicly and extensively self-described the specific model under discussion, in which case it appears in Part VI as a named case study and is quoted on its own claims.
• Every organogram presented as an example is a synthesis, meaning a representative geometry built by recombining structural features that were observed at multiple firms, with all identifying coordinates removed, so that no chart in this document corresponds to any single firm's team page.
• Statistics are reported as medians and ranges across the sample, and never as point values that could be traced to one firm.

Finally, because this dispatch introduces and reuses a number of technical terms, they are collected here in one place, grouped by what each set of terms describes. Each term receives its full formal definition at first use in the body of the dispatch; these tables exist so that a reader can check any term at any point without hunting for the section that defined it.

Series concepts. These terms describe the phenomenon under study and come from the AI Subsumption series itself.

The structural taxonomy. These terms describe the anatomy of the fund as reconstructed in Part II, and each weight class is defined separately because the classes behave differently at every stage of the analysis.

The minimum and the compression instruments. These terms are the analytical machinery of Parts III and IV.

The new institution. These terms describe what the compression creates.

The reconstruction encountered its first methodological obstacle immediately, and resolving that obstacle produced the analytical instrument on which everything else in this dispatch stands.

The obstacle is that nobody in African private capital means the same thing by the same job title. Across the 38 firms, the title of Principal describes, at different firms, a deal leader who holds board seats, a seniority band that is applied to the general counsel and to the chief financial officer, and the head of a founder-services platform. An Investment Director sits directly below the partners at one firm and below another Investment Director at the next. A Head of Investments is sometimes the equivalent of a partner and sometimes a delegated function head who sits three layers away from the investment committee. An Operating Partner is a full-time commercial operator at one firm and a fractional adviser at another. Meanwhile, a title that the industry's folklore treats as standard, namely the Associate Partner, the near-partner rung that supposedly sits between senior execution and the partnership, is effectively absent from all 38 firms. Two people with identical titles can hold entirely different authority, and two people with entirely different titles can hold identical authority. It follows that any analysis of this industry that counts titles, and most analyses of this industry do count titles, is measuring vocabulary rather than measuring institutions.

We do not treat this situation as a curiosity or as a complaint. It is the reason that a taxonomy had to be built, and the building of that taxonomy is how the archetypes of Part II were derived. Every observed title in the sample was sorted, cleaned, and mapped into a canonical structure in which the institutional role is the top-level node, defined by its decision rights, and the titles that African firms actually use hang beneath that node as observed instances. The complete sorting is presented in the two-column table below. The left column contains the canonical taxonomy node, which is defined by what the person actually does and decides. The right column lists the exact title strings under which the African firms in our sample label that same node.

One worked reading of the table will demonstrate the method. The canonical node in the second row, which is senior investment authority, is defined by a committee vote, by deal leadership, and by portfolio board seats. The African firms in our sample label that single node as Partner, as General Partner, as Senior Partner, as Investment Partner, and, in its executive usage, as Founding Partner. The same reading applies to every row of the table. It is also worth noting that several title strings appear in more than one row, among them Founding Partner, Operating Partner, Financial Analyst, and Head of Platform and Operations, and this repetition is itself a finding: the string alone can never locate the person, and only decision rights can do so. The table above is the clean output of the sorting exercise; the fully annotated version, complete with usage cautions, single-source markers, and cross-references, is set out as Annex C, and it serves as the reference vocabulary throughout this dispatch.

Once the taxonomy had been built, it yielded the study's first finding, which is that there is far more variation in titles than there is in actual organisational architecture. The delegated deal-leadership node alone carries seven different title strings across the sample, and those strings split almost perfectly along lines of institutional lineage: the word Principal is used in firms that descend from Silicon Valley venture practice, while Investment Director or Senior Investment Manager is used in firms that descend from development finance, and the split carries no structural information whatsoever. Meanwhile, the words that do carry structural information, meaning who votes on the committee, who signs the accounts, and who holds the board seat, appear nowhere in any title. Titles are local dialects, but institutions are not. Once the dialects are translated away, the institutions underneath turn out to be far more consistent, far leaner, and organised around a very different centre of gravity than the industry's self-image suggests.

The rest of this dispatch describes what that consistent institution looks like, establishes what its irreducible minimum is, examines what happens when the workflow inside it is subjected to the compression that Dispatch 01 described, and explains why the result of that analysis is not a forecast but a description of firms that already exist.

Before the African evidence is presented, the structure it will be compared against must be stated explicitly, because most readers carry it in their heads without examining it. The canonical investment firm of the developed markets, the firm of the textbooks, the recruitment brochures and the industry's self-image, is a pyramid. At its apex sit the partners or managing directors. Beneath them, each layer widens: principals or directors, then vice presidents, then associates, and at the base a broad cohort of analysts, hired in annual classes, who perform the industry's information processing and are simultaneously being trained by it. The pyramid is both a production engine and an apprenticeship system, and around it sits a heavily departmentalised institution: dedicated fundraising and investor-relations teams, in-house legal, tax and compliance departments, technology functions, human resources, and marketing.

The scale of this model at its fullest expression is a matter of public record. The largest listed alternative-asset managers, the houses of the KKR and Blackstone generation, publicly report headcounts that run to the thousands, and the largest conventional asset managers, of which BlackRock is the archetype, report headcounts in the tens of thousands. The model exists because the fee pools at that scale can fund it, and because, under the industrial logic that Dispatch 01 described, institutional capability and organisational mass were tightly linked: to be sophisticated was to be large.

This is the elephant that the Monday-morning operator of Dispatch 01 sat down to study, and its analyst pyramid is the layer that most commentary on AI and the investment industry assumes is about to be destroyed. The question Part II now answers from evidence is a prior one that the commentary skips: does the African fund actually have this shape at all? The answer, as the next three sections show, is that it does not, and the difference changes where subsumption lands.

When the dialects are stripped away, the 38 firms sort into three structural weight classes. Because these classes are used as an analytical instrument throughout the remainder of this dispatch, they must be defined precisely before they are used. The primary classification criterion is total headcount, and it is corroborated in every case by structure, because each band of headcount in the sample corresponds to a distinct organisational solution rather than to a larger or smaller version of the same solution. The classes are defined in the table below.

A note on the assignment convention behind the sample column is required. Firms whose documented headcount is a range that straddles a class boundary are assigned by their core payroll. One team that is embedded inside a global platform is classified by its standalone core, with the shared-services caveat recorded in the underlying research. Two legacy structures that were absorbed into larger platforms during the study period are classified as they stood before their mergers. The firm-level assignment key is held in the research appendix.

In full prose, the same three definitions read as follows. The large institutions are the control private equity houses and the multi-fund platforms, and they are the only firms in the sample that display genuinely elaborated internal structure, by which we mean multiple partners, specialised principals, in-house legal and finance departments, and dedicated risk, sustainability, and investor-relations functions. The mid-range firms are the modal African fund, comprising the institutional seed and early-stage venture firms, the growth investors, and the impact vehicles anchored by development finance institutions. This class contains most of the sample and most of the continent's active managers, and its structure is examined at full depth in Part III. The minimum-viable firms are the boutique partnerships, the single-founder vehicles, and the family and corporate capital arms. They are not immature versions of the mid-range firm; they are a distinct organisational solution, and understanding how they survive at that size turns out to be one of the keys to the whole analysis.

These three weight classes are not a descriptive convenience. They are carried through the remainder of this dispatch as three distinct analytical tracks, because the minimum viable structure, the location of the compression, and the destination institution all differ by class. Wherever the argument below states a single figure or describes a single structure, it is speaking of the mid-range modal fund, and the class overlays at each stage state how the minimum-viable and large classes diverge from it.

The headline statistics across the full sample, with the three multi-practice platforms excluded from the medians as outliers, are as follows.

Two of those rows are the empirical foundation of everything that follows, and they deserve to be stated as findings rather than left as statistics.

The first finding is that the analyst pyramid largely does not exist in African private capital. The standard mental model of an investment firm, in which partners sit at the apex above a widening base of junior analytical labour, describes almost nothing in this sample. Dedicated analysts appear at a quarter of the firms, in ones and twos, and they function as a capability rather than as a pool. The industry's junior information-processing layer, which is the layer that most AI commentary assumes is about to be destroyed, was never built here. This finding bears directly on the series, because the Monday-morning operator of Dispatch 01 began by studying private equity's analyst pyramids, and the first empirical result of mapping the African elephant is that this particular part of the elephant is missing. African funds skipped the pyramid for reasons of fee economics rather than foresight, and that accident turns out to matter enormously for where subsumption actually lands.

The second finding is that the largest concentration of institutional mass is the operating spine. Finance, fund operations, reporting, compliance, investor relations, platform services, and administration together account for one third to one half of headcount at nearly every firm, and the spine is the only organisational feature that is present at all 38 firms. At the control private equity houses, the spine is the outright majority of the institution; in one reconstructed structure, ten of seventeen people sit entirely outside the deal chain. Whatever artificial intelligence does to the investment industry, the mass that it will actually encounter is not a population of analysts. It is the spine.

A third structural finding completes the anatomy. In every firm where reporting relationships could be verified, the organisation is not one vertical hierarchy but three short chains sharing one apex: an investment chain, a portfolio and platform chain, and a fiduciary and operating chain, all of which branch directly beneath the managing partner. Nothing in the sample shows the finance function or the platform function reporting upward through the investment team. Half of what looks like the middle of an African fund is not middle at all; it is sideways.

When they are translated out of dialect, the 38 firms resolve into seven recurring organisational forms. For each form, this section gives the composite geometry and, where the anonymisation protocol permits, synthesised examples of the structure as it was actually observed. Two of the archetypes are presented at pattern level only, because so few firms exhibit them that any organogram would identify its source.

Five findings emerge from the reconstruction, before a single word has been said about technology.

First, the analyst pyramid does not exist. Africa's funds never hired the junior information-processing layer that dominates the industry's self-image elsewhere.

Second, the operating spine is the institution's largest mass and its only universal feature. This holds from the four-person boutique, which rents its spine, to the forty-person control house, which is mostly made of spine.

Third, the fund is three short chains sharing one apex rather than a pyramid, and half of its apparent middle is in fact lateral.

Fourth, capital structure determines which functions exist, and it does so more strongly than either size or strategy. We call this the DFI floor.

Fifth, the visible firm is not the whole institution. Around every payroll sits an invisible organisation of external administrators, auditors, counsel, fractional venture partners, non-executive committee members, and, in the case of embedded teams, the shared services of global platforms. A four-person fund can be a ten-seat institution, and a twelve-person fund can contain three deciders. Headcount, like titles, measures the wrong thing.

Those five findings hold across the whole sample. Beneath them, the three weight classes each carry findings and anomalies of their own, and they are stated here class by class, in skimmable form, so that the anatomy is complete before the argument moves on.

• Partner-titled seats form the majority of the payroll, which is the inverse of every other class.
• Not a single dedicated analyst exists anywhere in the class.
• Judgment is systematically rented rather than salaried: fractional venture partners appear throughout, at more than one firm with formal investment-committee votes, and at the smallest end of the class the investment committee is externalised entirely to non-executive seats.
• The operating spine sits wholly outside the payroll.
• The family and proprietary-capital vehicles in this class delete whole functions that the textbook treats as universal, with fundraising and investor relations foremost among them.
• The class produces the sample's boundary-dissolving anomalies, meaning seats in which employment status, capital provision and governance authority separate from one another, so that full-time-equivalent arithmetic simply fails; one firm's three-partner core sits above separate country advisory companies, replacing headcount with entity geometry altogether.

The implication of these findings is that most of this class already operates at or near the minimum institutional mass, which is why, as Part IV will show, the compression has almost nothing to remove from it.

• This is the modal class, and it is where the full three-chain geometry appears in its complete form.
• The title dialect split runs straight through it, with the Principal family of titles in firms of one lineage and the Investment Director and Investment Manager family in the other, for what is the identical institutional layer.
• The DFI floor binds precisely here: no DFI-backed firm in the sample operates below roughly nine or ten people, which is why the floor of the class sits where it does.
• The combined seats concentrate in this class, exemplified by the executive assistant who is also the investor-relations function, and by the chief of staff who carries investor relations, reporting and firm operations in one role description.
• The class contains the sample's productivity anomaly, which is a team in the low teens carrying a portfolio an order of magnitude larger than its own headcount on an explicit coverage model.
• The class also contains the sample's structural anomaly, which is a firm whose entire professional middle band is deployed inside portfolio companies rather than at the centre.
• Both consolidation casualties of the study period were sub-scale members of this class.

The implication of these findings is that this class carries the three additions that fee income exists to fund and that compression will target, and that its lower boundary is set by covenant rather than by choice; the last finding is also the empirical basis of the trap described in Part III.

• The operating spine is not merely the largest mass but the outright majority of headcount.
• This is the only class in which in-house general counsel, a chief risk officer, a dedicated head of investor relations or multi-seat finance exist as separate positions.
• The grade-not-role compound titles, meaning a Principal who is the general counsel and a Principal who is the chief financial officer, are observed only in this class, because only here are the functions large enough to be graded into the investment band.
• The deal chain is proportionally at its thinnest, at five to seven people inside institutions of seventeen or more.
• The class contains the two platform outliers whose headcount scales with programme delivery rather than with analysis, and those two firms are excluded from every median in this dispatch for that reason.

The implication of these findings is that the large firm's mass is accountability-dense rather than workflow-dense, which is why, as Part IV will show, it experiences the smallest proportional compression of any class.

These findings frame the question that the next part answers. If so much of the institution is shell, spine, and rented judgment, then what is the irreducible core? Let us forget artificial intelligence entirely, strip everything away, and ask what must remain.

The minimum viable fund is usually discussed as if it were a single question, but it is actually four questions with four different answers. What does regulation require? What do limited partners require? What do auditors require? And what actually happens in practice? The reconstruction allows us to answer each of these from evidence.

Regulation requires a licensed or authorised manager entity with accountable officers, which is to say directors, a compliance arrangement, and audited accounts. In the adviser-model structures described in Part II, regulation locates decision authority in the regulated entity regardless of where the people physically sit. What regulation nowhere requires, in any jurisdiction touched by this sample, is a particular headcount. Regulation names accountabilities; it does not name staff plans.

Limited partners, and above all the development finance institutions, require independent fund administration, a compliance owner, an ESG function, key-person provisions, defined reporting cadences, and, structurally, the segregation of investment decision-making from valuation and from the movement of money. These requirements arrive by covenant and by side letter. They are the reason that the DFI floor exists, and they too are written as functions and named officers rather than as headcount.

Auditors require that the books are maintained independently of the deal team, that valuations rest on a process with challengeable inputs, and that net asset value is verified by a party outside the investment function. This is the deep reason why the operating spine cannot simply be deleted by anyone, whether human or machine: parts of the spine exist precisely in order to be independent of the people who would benefit from deleting them. In the reconstruction, this requirement appears as a recurring triangle consisting of an in-house finance owner, an external administrator, and an external auditor. At one growth firm in the sample, the arrangement is stated in a published role description in so many words: the chief financial officer's job is defined as the interface between the investment team and the external fund administrator. Ownership sits inside, execution sits outside, and verification is independent. The triangle is the segregation of duties, achieved by institutional geometry rather than by internal staffing.

Practice, finally, shows that within all of this scaffolding, two to four people actually decide. Everything else in the institution supports those people, verifies their outputs, or reports on the results.

Before the minimum is drawn as a structure, its logic can be stated as a table. Each row names an irreducible function, explains why it exists, and records whether the sample shows it being carried outside the payroll.

When the four requirement stacks of Section 7 are put together with this logic, the minimum viable institutional structure of an African fund, with no artificial intelligence anywhere in the picture, is the following.

This structure places four people on the payroll, and those four people occupy an institution of roughly ten accountable seats once the external shell and the independent governance are counted. Where DFI capital is present, an ESG seat and often a compliance seat must be added, which raises the payroll floor to about six people. The sample confirms this arithmetic from both directions: the smallest credible firms in the study operate with four to six people surrounded by exactly this shell, while no DFI-backed firm in the sample operates below roughly nine or ten people in total.

The minimum must now be read against the three weight classes, because the minimum viable institutional structure is not a target towards which every fund should shrink. It is, rather, the invariant core that all three classes demonstrably contain, and what distinguishes the classes from one another is what each of them adds to that core and where each of them locates the verification shell.

The minimum-viable class simply is the minimum viable institutional structure. It places three to five people on the payroll, it rents part of its judgment through venture partners and non-executive committee seats, and it keeps the entire shell external. There is nothing in this class to strip away, because nothing beyond the core was ever built.

The mid-range class is the minimum viable structure plus three additions, which are a delegated deal-leadership layer with an execution band beneath it, a platform function, and in-house ownership of the operating spine, while the administrator and the auditor remain external. This produces a firm of nine to thirteen people, rising by one or two where the DFI floor applies. These three additions are what fee income at this scale exists to fund, and, as Part IV will show, they are also precisely where the compression lands.

The large class is the minimum viable structure plus internalisation. Control private equity does not merely add layers; it moves parts of the verification shell onto the payroll, in the form of in-house counsel, multi-seat finance, a formal valuation apparatus, and, at the largest firms, dedicated risk, because control transactions demand these capabilities. The result is a firm of seventeen people and upward, the majority of whom are institution rather than deal chain.

The structure of the argument can therefore be summarised as one core wearing three different coatings, and that image should be kept in view, because the compression analysis that follows behaves entirely differently depending on which coating it strikes.

It is also worth stating clearly what this minimum is not. It is not two partners and nothing else. The fantasy version of the lean fund, which is the version that appears whenever this subject is discussed casually, deletes the control seat and the external shell, and in doing so it stops being an investable institution, because it retains no segregation of duties, no independent verification, no accountable finance ownership, and nothing on which an institutional limited partner can rely. The boutiques in our sample survive at four people precisely because they never deleted any of those things. They moved them off the payroll, and they kept every accountability intact.

The pre-AI minimum has a price, and that price has been quietly restructuring the African fund industry.

A core of four to six people, plus an external shell, plus, where DFIs are present, a compliance and reporting apparatus, costs real money, and the management fees on a small fund cannot always carry that cost. Somewhere below a threshold of assets under management, a standalone manager cannot fund its own spine, and the sample suggests that in DFI-facing markets this threshold sits high.

The sample contains the consequence of this arithmetic twice over. Within the study period, two established managers in the same region ceased to exist as independent firms, and each of them folded its funds and its team into a larger platform. In both cases the publicly stated logic was the same: at their scale, executing a regional strategy while carrying an institutional-grade operating apparatus was not economic, and merging meant sharing someone else's spine. This is the sub-scale manager's trap, and until now it has had exactly two exits: a manager could stay small and starve its institution, or it could consolidate and disappear.

The reader should hold that trap in mind, because it is about to become the hinge of the entire analysis. Everything in this Part was established with no reference to technology. The minimum viable African fund, as described above, exists today, and it has existed for years. No artificial intelligence was involved in reaching it.

Now, and only now, do we introduce AI, because it is now finally clear what would actually be compressed.

The question that artificial intelligence poses to an investment firm is routinely asked in the wrong unit. It is asked about jobs, in the form of the question of which jobs disappear. But Parts II and III have established that a fund is not a collection of jobs. It is a governance system for allocating capital, it is made of functions, and every function in it blends, in different proportions, six distinct things. Because the compression analysis rests on separating these six things from one another, each of them must be defined explicitly before any scoring begins. Workflow is the routine production of information, and it covers screening, drafting, assembling, reconciling, and reporting. Judgment is the exercise of discretion under uncertainty, and it answers questions such as which company, which founder, which price, and which intervention. Governance is participation in the structures through which decisions are legitimised, meaning committees, boards, and approval processes. Fiduciary duty is the legal obligation to act in the interests of the capital's owners, and it attaches to specific named officers. Relationships are the trust held with founders, limited partners, co-investors, and regulators, and this trust cannot be reassigned by an organisation chart. Accountability is the property of being the person to whom a failure is attributed, and it lives in the signature, the officership, and the opinion.

When every role in the 38-firm reconstruction is classified along those six dimensions, a single pattern dominates the result: workflow intensity and accountability intensity are close to inversely correlated. The people who do the most routine work, meaning the analysts, the associates, the fund-operations staff, the reporting producers, and the administrators, carry the least institutional responsibility. The people who carry the most responsibility, meaning the managing partners, the committee members, the finance owners, and the compliance officers, do the least routine work. Work being done and responsibility being carried are therefore not the same thing, and in the anatomy of a fund they largely live in different seats.

That inverse relationship is the compression map. It tells us in advance where automation will bite and where it will not, and it converts the vague question of what AI does to funds into three precise questions. Which parts of the institution are workflow? Which parts are judgment? Which parts are accountability?

The analysis in this Part is governed by a single rule, which is applied before any enthusiasm for the technology is allowed into the room:

Any function involving legal accountability, fiduciary responsibility, regulatory obligation, investment authority, valuation sign-off, LP trust, or segregation of duties is presumed non-subsumable until evidence demonstrates otherwise.

This rule is not caution for its own sake. It follows from what Part III established, which is that those functions are the institution's reason to exist. A fund is a promise structure with a workflow attached, and the analysis must be able to tell the two apart.

When every function in the reconstructed fund is run through that constraint, and tested against what current AI systems demonstrably do, the functions sort into three buckets. This three-way classification is the first of two instruments that this Part introduces, and it operates at the level of the function itself; the second instrument, which is the set of four operating modes introduced in the next section, operates at the level of how a surviving function is performed. The boundary between the three buckets is not technical difficulty. It is the question of who relies on the function.

The functions that disappear are convenience functions, which are performed for the fund's own benefit. They include screening inbound flow, producing research, drafting memos and market maps, assembling reports, and coordinating documents, calendars, and correspondence. Nobody outside the firm relies on how these things were done, and they therefore convert to software with no institutional residue. In the vocabulary of Dispatch 01, these functions are the fund's subsumption zones: they are the layers that exist primarily because older organisational systems required humans to move information between nodes.

The functions that survive are reliance functions, which are performed so that someone else can trust the result. They include investment authority, LP and owner trust, portfolio board participation, valuation sign-off, governance, compliance ownership, independent administration, and audit. The purpose of these functions was never the labour involved in them; their purpose was the accountable party attached to the labour, and software cannot be that party.

The functions that become infrastructure are the information-processing middle of the institution. They include sourcing, monitoring, knowledge management, reporting workflows, and portfolio intelligence. These functions neither vanish nor stay human. They convert from salary to software, and in converting they change category on the balance sheet of the institution: they move from being a cost line to being an owned institutional asset, which is maintained and versioned like any other proprietary capability and which, increasingly, forms part of what limited partners are underwriting when they back the manager.

To make the framework concrete, we take the modal institution from Part II, which is the mid-range archetype-B firm, the institutional seed and early-stage VC, of roughly eleven people, consisting of two partners, a delegated deal lead, three execution staff, a platform head with one support seat, a finance owner with fund-operations and accounting support, and a combined administration and investor-relations seat. We then run each of its functions through the three buckets, this time carrying the exact titles that hold each function today, drawn from the roster in Annex C, and classifying the post-compression operating mode of whatever survives.

Before the table of functions is presented, the operating-mode classification itself must be introduced, because it is used throughout the remainder of this dispatch. Every function that survives compression is performed, afterwards, by some division of labour between the machine and a human, and that division takes exactly four stable forms in this analysis. The four modes are defined in the table below.

The distinction between the second and third modes matters and should not be blurred: in an AI-led function the machine is the producer and the human is the guarantor, whereas in a human-led function the human is the producer and the machine is the amplifier.

The status markers introduced in the opening dashboard are used in the tables that follow: the seat or act remains human; the machine produces and a human owns; the dedicated seat is subsumed; a newly created seat.

The functions describe the institution, but people occupy titles rather than functions, and so the same analysis must be run once more in the other direction. Title by title, across the roster of Annex C, the following crosswalk states what vanishes as a dedicated seat, what remains, and in which of the four modes the surviving seat operates.

The title crosswalk

When the crosswalk is read as a whole, four role futures emerge, and they map exactly onto the four modes. The seats that vanish outright are the AI-only rows, meaning the analyst, the associate-production seats, the platform support, the communications and administration seats, and the reporting producers, and the African irony is that most of these seats were never widely occupied here in the first place. The seats that remain under AI-led operation are the owners of machine production, meaning the control owner, the compliance officer, the platform owner, and the merged chief-of-staff functions. The seats that remain human-led are the relationship and coverage owners, meaning the deal leads, the portfolio managers, the embedded operators, counsel, and talent. The only-human seats are the institution itself, meaning the partners, the committee, the sign-offs, the opinions, and the independent governance. One further category cuts across the modes rather than sitting inside them, and that is the set of institutionally required separations: fund administration, valuation review, audit liaison, and capital-call and distribution verification all remain even at full automation, because their entire purpose is to be independent of the people, and now of the systems, that produce the numbers. Automating a verification function does not remove it. It changes how the verifier works.

The class overlay. The function table and the crosswalk above are drawn on the mid-range reference fund, and they must not be read as if the bifurcations of Part II had disappeared, because the same compression strikes the three weight classes in three different places.

The overlay explains an asymmetry that a single-track reading would miss entirely. The class with the most subsumable titles on its payroll, which is the mid-range, is the class that the technology restructures; the class below it multiplies, and the class above it trims. One technology produces three different fates, and the fate is determined by which coating of the minimum viable core the technology strikes.

Two findings from this analysis are strong enough to discipline every forecast in this genre, including our own first-stage conclusions, which these findings corrected.

The first finding is that, across every role at 38 firms and under every regulatory regime in which those firms operate, the capacity of AI to absorb workflow varies enormously, while its capacity to absorb accountability is zero, everywhere. No regime in the sample permits software to hold a directorship, a compliance officership, a valuation sign-off, or an audit opinion, and no LP covenant names a model as a key person. Early versions of this research assumed that the mid-band of the fund, meaning the five to eight execution and operations seats, could simply be eliminated. When that assumption was tested function by function, it failed. What can be eliminated is dedicated full-time labour; what cannot be eliminated is a single accountable function. The unit of AI subsumption is the full-time employee, not the role, and never the accountability.

The second finding closes the loop from the other side, because it answers the obvious objection that some firms will surely just remain human. In the classification of organisational forms that this research produced, the category of funds that rationally operate with no AI at all is empty, and it is empty not because the technology conquers every firm but because adoption is ambient rather than elective. The external shell automates on its own schedule regardless of what any fund decides: administrators, counsel, and audit firms are converting their own production to software for their own commercial reasons, and platform providers now sell agentic fund operations to emerging managers as a subscription. A fund cannot opt out of its administrator's automation. The only decision that is actually available to a manager is whether institutional compression is designed or merely absorbed.

When the results of Part IV are applied to the anatomy of Part II, each archetype can be redrawn. The institutions are the same, the accountable seats are the same, and the payrolls are post-compression.

The tables of Part IV state what vanishes and what remains, but they are incomplete without a third category, which is the set of seats that the compression creates. The AI-native fund is not merely the old fund minus its subsumed labour. The genuinely new elements are summarised in the table below and then examined in turn, because each of them raises a question that the industry has barely begun to answer.

The first new seat, the architect-operator, deserves elaboration, because it is already visible in the field in three different sizes. At one of the firms named later in this Part, the managing partner personally built the firm's fifteen-agent operating system and remains its architect. At another named firm, the seat exists as a dedicated applied-AI adviser who works alongside the investment partners. At the largest AI-native firm, the seat has become an entire engineering department. Whether the person carrying this seat is titled data scientist, AI engineer, head of platform, or something not yet invented matters as little as any other title in this dispatch. The decision rights are what define the seat: this person owns the infrastructure in the way that the chief financial officer owns the books.

The second element, agent supervision, exposes the requirement on which the whole structure rests. Every AI-led function in the tables of Part IV has a human owner who reviews and signs what the machine produces, which means that the deal lead is now also the supervisor of the drafting and diligence agents, that the control owner is the supervisor of the accounting and reporting pipeline, and that the compliance officer is the supervisor of the monitoring agents that feed the compliance calendar. Supervision is domain work rather than technology work: the architect-operator keeps the agents running, but only the domain owner can judge whether their output is right. A person cannot supervise an agent performing a craft that the person has never practised. The control owner who audits a fund-accounting pipeline must know fund accounting. The deal lead who corrects a machine-drafted memo must know what a defensible memo contains. The compliance officer who challenges an automated screening run must know what the run should have caught. The supervising human must, in short, be a master of the subsumed craft.

This raises the third question, which is where the masters come from and who verifies them. The first generation answers itself. Today's supervisors are veterans of the manual system, meaning former analysts, fund accountants, controllers and audit-trained professionals, who acquired their pattern recognition by doing, for years, precisely the work that the machine now does, and the external shell disciplines them exactly as it always did, because the auditor challenges the accountable human owner rather than the software, and the audit cycle therefore remains the institution's teacher of last resort. The second generation, however, is an open problem, because the training infrastructure was the work itself. Dispatch 01 named this the apprentice problem: drudgery was also apprenticeship, and the pipeline that produced masters runs through exactly the layers that AI compresses first. The evidence in this dispatch sharpens the problem and, unexpectedly, softens it for this continent. Africa's funds, as Part II established, never built the analyst pyramid, which means that African investment judgment was never actually formed through pyramid apprenticeship in the first place. It was formed through early proximity to partners, early deal exposure and early responsibility, and that is precisely the training geometry that survives compression intact. The continent's accidental head start therefore extends one layer further than Part VII will claim: not only is there no legacy layer to unwind, but there is also no legacy training pipeline to mourn. What remains genuinely unresolved, in Africa as everywhere else, is the supply of the control-side masters, meaning the people who can supervise the accounting, valuation and compliance pipelines once the junior seats of those professions thin out. The honest position, which this dispatch adopts from the series, is that this question has been identified but has not yet been answered.

We now return to the trap that closed Part III. The sub-scale African manager's dilemma was that the fixed cost of the spine set a floor on viable assets under management, and below that floor the available choices were starvation or merger.

Institutional compression attacks that floor directly, and this is the largest single claim in this dispatch: AI changes the minimum viable AUM of a fund manager. If the operating spine converts from several salaries into software plus one accountable owner, then the fee base that is required to sustain an institutional-grade manager falls by something like half. The consequences of that fall are not staffing consequences. They are market-structure consequences.

The consequences follow one another directly.

• More managers become viable at smaller fund sizes, because the fee base required to carry an institutional-grade spine has fallen.
• First-time manager formation, meaning a first fund raised by a new general partner, becomes structurally cheaper, in an industry whose gatekeepers have spent a decade telling African managers that they are too small.
• The consolidation logic that has been quietly thinning the continent's mid-tier loses its engine, because the spine-sharing merger stops being the only exit from sub-scale operation once a compressed spine no longer needs to be shared.
• The DFI reporting economy, meaning the heavy documentary apparatus that development capital installs, becomes cheap to serve, which removes one of the quiet deterrents to taking development money at small fund sizes.

For a continent whose fund industry has been repeatedly told that it has too many small managers, the arithmetic of what counts as too small has just moved. That is not an efficiency observation. It is a market-structure observation.

Everything up to this point could be read as a projection, and a careful, empirically grounded projection at that, but a projection nonetheless. It is not one. The structures described in Part V already exist. This Part locates them, and it begins by stating plainly where they do not yet exist, which is in the sample itself.

As of the time of writing this dispatch, we identified no self-described AI-native private equity or venture capital firm within African private capital, and none within our 38-firm sample.

This does not imply an absence of AI adoption, and the evidence for adoption, though anecdotal, is visible in the firms' own publishing. The public posts flowing from the sample firms' company pages increasingly carry the stylistic fingerprints of machine drafting, the em-dash prominent among them, which is a punctuation habit this industry rarely typed before 2023. From that observable telltale we infer, and we label this explicitly as an inference from public behaviour rather than a verified finding, that firms across the sample are using AI for writing purposes: drafting emails, producing investor reports, preparing communications and presentations, and supporting research. Anecdotally, the frontier of adoption is marked by the engagement we observe with practitioner content, of the kind exemplified by guides on how to use AI assistants for financial modelling; commenting on such a guide is, for much of the sample, the visible extent of the journey so far.

This matters because using AI and being AI-native are materially different things, and the distinction is critical to everything this dispatch has argued. The difference is stated in the table below.

A firm in the left column has changed its stationery. A firm in the right column has changed its institutional geometry. Every firm in our sample that shows evidence of AI at all sits, as far as public evidence reveals, in the left column. The subsumption zones identified in Part IV remain, in African private capital, almost entirely unclaimed.

A misreading must be blocked before the field is surveyed, because much of the current AI discussion implies it accidentally: the idea that an AI-native firm is a firm that built its own AI. That idea is nonsense, and the equivalent claim in an earlier era makes the nonsense audible: it is the claim that a software-native company is one that built its own internet. It is not. Most firms consume infrastructure that others built, and the AI-native firm is no exception.

For the avoidance of any doubt, AI-native does not mean any of the following.

• It does not mean training foundation models.
• It does not mean building frontier large language models.
• It does not mean designing proprietary chips or operating GPU clusters.
• It does not mean running hyperscale datacentres.
• It does not mean competing with the frontier laboratories, the OpenAI, Anthropic, Google and xAI tier, whose entire business is the models themselves.

As of 2026, virtually no investment firm, advisory firm, law firm or private equity firm on earth is doing any of these things, because the economics are prohibitive: frontier model development is a multi-hundred-million-dollar undertaking that belongs to a handful of laboratories and the capital markets behind them. Every AI-native firm named in this dispatch, and OFP itself, rents that layer.

The interesting question about an AI-native firm is therefore not what model it is building, because the answer is none. The interesting question is what institutional functions it is orchestrating on top of models built by others. The working stack of the category, verifiable from the firms' own descriptions and from the broader 2026 tooling landscape, has settled into recognisable layers.

Two observations about the stack matter for everything else in this Part. The first is that every layer of it is rented or assembled, and none of it requires building a model; the capital cost of becoming AI-native is measured in subscriptions and engineering time, not in datacentres, which is precisely why Part V's economics hold at African fund sizes. The second is that the stack is what the architect-operator of Section 17 actually owns and operates: that seat is an assembler and orchestrator of this stack, not a model researcher.

With the stack in view, the distinction that Section 19 introduced can now be drawn precisely, as two different flows through the same tools.

In the adoption flow, which describes the typical African private capital firm today, frontier assistants and a document copilot are pointed at existing outputs: emails, reports, presentations and research get faster, and the institution stays structurally identical. In the AI-native flow, the same rented models are wired downward through knowledge systems and workflow orchestration into the institution itself, so that the organisation is designed around AI-enabled workflows rather than merely accelerated by AI-enabled typing.

Behaviour across the whole professional landscape now sorts into four adoption archetypes. These adoption archetypes classify behaviour, and they must not be confused with the organisational form archetypes A to G of Part II, which classify structure; a firm has one of each.

The distinction matters because, without it, readers will reason as follows: Kadan uses Claude, therefore Kadan is AI-native. That inference is wrong in both directions. A firm can use AI heavily and remain structurally identical to its pre-AI self, and by the usage definition most of the professional world became AI-native at some point in 2024, which is exactly why the usage definition is useless. The real test is structural: have sourcing, diligence, reporting, knowledge management, portfolio monitoring and internal operations actually been redesigned around AI-enabled workflows, with the headcount derived from the redesign, as the design sequence of Part VII will formalise?

The defining feature of an AI-native investment firm is not the model it uses. It is the extent to which institutional workflows have been redesigned around AI-enabled execution. That single sentence is the conceptual hinge of this entire Part, and every case that follows should be read against it.

Before the fund cases are presented, one adjacent category deserves attention, because it is where AI-native institutional design first appeared in professional services, and because it explains a constraint on this dispatch's own evidence.

A field of self-described AI-native advisory and professional-services firms has emerged since roughly 2024. These are system players rather than private-capital specialists: they are advisory, consulting and institutional-design firms for which private equity is one domain among the many they advise on. The category is verifiable and growing.

• A Toronto consultancy describes itself as a canonical AI-native consulting firm, founded on the model in 2024 and staffed overwhelmingly with engineers rather than analysts.
• An Australian consultancy has publicly documented building itself AI-native from day one in 2025.
• A market map compiled in 2026 counts more than two hundred self-described AI-native service firms across some seventy industries.
• The world's largest AI laboratories have begun backing AI-native enterprise-services ventures directly.
• Odit Frontier Partners itself belongs to this category, having self-described as AI-native since 2025, with capital architecture as its domain.

The constraint this category imposes on evidence, however, must be stated honestly. Firms of this kind, OFP included, keep their methods proprietary. The internal systems are the competitive asset, and they are therefore a black box: no external observer can verify the architecture, and the firms do not disclose it. For exactly this reason, adjacent AI-native firms cannot serve as structural case studies in a dispatch that has committed itself to evidence discipline, and this dispatch does not speculate about any firm whose methods are not public. The consequence is unavoidable: for verifiable, structurally documented cases of AI-native institutional design in private capital, we can only borrow from global examples, and that is what the remainder of this Part does.

Secha Capital, a Southern African growth-capital firm, has publicly and extensively described an organisational model that it calls the Operator-Investor. The firm's centre consists of its three co-founders. Its entire professional middle band, which comprises six operators drawn from strategy consulting, finance, and operating backgrounds, is deployed inside its portfolio companies, where each operator is embedded full-time for around a year in order to execute the firm's value-creation framework alongside the founders it backs. The firm describes this model, in its own words, as human capital arbitrage, and it has extended the model into a track that places operators directly into portfolio C-suites.

Read against the framework of this dispatch, Secha is the pre-AI existence proof of the entire geometry. Its execution capacity does not live at the centre of the institution; it lives in the portfolio. The centre holds exactly what Part IV says cannot be compressed, namely judgment, committee authority, capital relationships, and governance. The firm reached capability without central headcount with no artificial intelligence whatsoever, by reorganising people instead of deploying software. This demonstrates something that the technology debate persistently misses: the destination is an organisational design, and software is only one road to it.

Kadan Capital, a Singapore-based early-stage firm founded in 2024, publicly describes itself as an AI-native venture firm. Its founding partner has written that the firm built, internally, a machine-learning capability that would have cost seven figures in data-science staffing in the era before large language models; that administrative work which would otherwise occupy multiple full-time employees now runs at a token cost that is a fraction of one junior hire; and that the automation covers thesis research, founder-history diligence before first meetings, memo drafting, and document work, with the explicit purpose of freeing the partners' human hours for time with founders.

The firm's structure and its disclosed AI layer, with each element carrying its evidence status, are as follows.

The AI infrastructure layer, as self-described by the firm, comprises machine-learning-assisted sourcing, automated thesis research, founder-history diligence performed before first meetings, memo drafting, and administrative work run at token cost. No element of this layer has been externally audited, and this dispatch quotes all of it as the firm's own claims.

It is worth noticing what the claims describe, and equally what they do not describe. They describe the conversion of the execution band and the administrative spine into an owned software layer, which is precisely the set of functions that Part IV places in the disappear and become-infrastructure buckets. They do not claim partner replacement, committee automation, or any transfer of accountability, which is precisely the set of functions that Part IV marks as non-subsumable. A firm with every incentive to overclaim draws its own line exactly where the institutional analysis draws it.

It is also worth stating what Kadan is not, because the point is routinely missed. It is not three people and a chatbot. It is three to four humans of judgment, an owned AI infrastructure asset, and an unchanged external shell of administrators, auditors, and counsel around them. The visible payroll was always going to be three or four people at that fund size. What has changed is the capability density per human, together with the fact that part of the institution is now software that the firm owns.

The two cases can be set side by side as follows.

Here are two firms, on two continents, using two technologies, arriving at one geometry. The middle of the institution, meaning the layer between judgment and verification, is where compression happens, however it is achieved. The judgment core and the verification shell are where compression stops, whoever attempts it.

The two developed cases are not isolated, and the field forming around them can be documented firm by firm, applying the same evidence discipline: for each firm, the public AI-native claim is quoted as a claim, the human structure is stated only to the extent that it is disclosed, and the AI layer includes only what the firm itself has described or what credible external reporting has recorded. Where something is unknown, the tables say so.

The firm's founder has publicly described an internal operating system, which the firm calls Ripple OS, built around roughly fifteen AI agents, and he states that it lets the firm's four-person team do the work of up to a hundred people. That figure is a founder claim reported by the press, and the same reporting carries his candid admission that the partners have never been busier, which is to say that capacity has been multiplied rather than leisure created.

The firm has described itself as an AI-native quantitative venture firm since its founding in 2017, which makes it the longest-running self-declaration in the field, and it built its stack before it scaled its fund. It advertises small seed cheques with soft commitments in days rather than months.

The firm is publicly described, and describes itself, as data-driven to the point of machine-reviewing tens of thousands of companies each week, a throughput it estimates at hundreds of times the industry average, with a small and deliberately engineer-heavy team.

SignalFire is the largest firm publicly marketing itself as AI-native. It manages roughly three billion dollars, a figure verified by external reporting of its fundraising, it employs a large team that is heavy with engineers, and it describes a proprietary platform, which it calls Beacon, as tracking hundreds of millions of professionals and tens of millions of organisations. It did not shrink; it rebuilt the analyst pyramid as an engineering department. This reveals what actually defines the category: AI-native does not mean small. It means that the design sequence ran in the right order, with process and infrastructure first and headcount derived afterwards. Kadan derived a small firm from that sequence, SignalFire derived a large one, and Secha derived its humans from a human-capital process. The sequence, and not the headcount, is the signature.

Below the level of individual firms, one further datapoint completes the field. Platform providers now sell agentic fund operations, covering sourcing, diligence memos, LP management and quarterly reporting, to emerging managers as a subscription, and they market the proposition in exactly the terms of this dispatch, promising a solo GP the leverage of a ten-person fund. The spine has become a product, and this is the mechanism through which, as Section 15 argued, adoption becomes ambient rather than elective.

Three misreadings will attach themselves to this analysis, and it is cheaper to refuse them here than to correct them later.

Every structure described in Part V has a currently operating counterpart named in this Part. The compressed fund is not the next era of the industry; it is a present-tense minority position whose economics are visibly better.

Nothing in 38 reconstructions, and nothing in the public claims of any AI-native firm, however self-interested those claims may be, supports the substitution of investment judgment, committee authority, or LP trust. The firms that are furthest into the technology are the most explicit that human hours have moved towards founders rather than away from them.

The lean fund that deletes its control seat, its administrator, its auditor, and its independent governance is not an AI-native institution; it is an uninvestable one. Every credible compressed structure in this dispatch keeps the full set of accountable seats and the full external verification shell. The compression is real, and it happens strictly inside the promise structure, never to the promise structure itself.

Underneath every finding in this dispatch sits one change that is deeper than any organisation chart: the order of operations by which an investment firm comes into existence has reversed.

In the old sequence, headcount is an input: the firm hires the institution and then discovers what it does. In the new sequence, headcount is an output: the institution is conceived as a set of functions, the functions are built and automated, and the people who remain are exactly those whom the analysis of Parts III and IV says must remain, namely the judgment core and the control owners. Design begins from the minimum viable institutional structure, and it adds nothing that the promise structure does not require.

For African managers, this inversion carries an advantage that is rarely available to this continent's industries: an incumbency problem that belongs to somebody else. Africa's funds never built the analyst pyramid, so there is no legacy layer to unwind and no threatened middle whose defence would distort the redesign. The modal African fund, as Part II showed, was already three short chains around a small apex, was already renting half of its institution, and was already compressed by scarcity into something close to the right shape. A first-time African GP raising a fund today can run the AI-native sequence from day one, at a minimum viable scale that the old economics never permitted, in an industry whose consolidation trap has just lost its engine. For once, having built lean out of necessity converts directly into having built correctly for what comes next.

We can now assemble everything: the reconstruction of Part II, the minimum of Part III, the compression of Part IV, the redrawn archetypes of Part V, and the operating proof of Part VI. The fund after headcount resolves to a stable and specific form, which consists of the following.

This institution carries roughly the same number of accountable seats as its pre-AI predecessor, roughly half the payroll in the mid-range and above, and two to four times the institutional capability per human. It also carries one genuinely new thing, which is a proprietary infrastructure layer that limited partners are, in effect, underwriting alongside the team, and this means that manager selection itself begins to include an assessment of software.

Stated one final time by weight class, so that the model carries the bifurcations on which it was built, the result is as follows.

Across the three classes there is one invariant: in every row, the judgment core, the control ownership, the owned software layer, and the external verification shell are all present. What varies is only how much coating surrounds the core, and how much of that coating the machine has taken.

Funds are governance systems for allocating capital. Artificial intelligence compresses the labour inside the system, but it does not, and within any current legal or fiduciary regime it cannot, eliminate the promises that make the system investable. The managers who understand the difference between those two sentences will be the credible institutions of the next cycle. The managers who confuse them, in either direction, whether by defending headcount as if it were the institution or by deleting accountability as if it were headcount, will not.

The conclusion should return to what this research actually did and actually found. We reconstructed the African investment fund from public evidence, we identified its minimum viable institutional structure, and only then did we examine what happens to that structure when workflow becomes abundant. The core finding has held at every step: AI compresses labour, and it does not eliminate institutions. It follows that the future fund is not defined by fewer people. It is defined by a different relationship between judgment, accountability and infrastructure, in which judgment remains scarce and human, accountability remains legally and fiduciarily non-transferable, and infrastructure absorbs the labour that once sat between them.

None of this is confined to venture capital. The anatomy that this dispatch has reconstructed, which consists of a small judgment core, a verification shell, and, between them, an operating spine that exists primarily to move information from where it is produced to where it is relied upon, describes consultancies, accelerators, family offices, fund-of-funds programmes, and development institutions with equal precision. Wherever that anatomy holds, the same three-part fate awaits the same three parts of the institution. Dispatch 01 closed by naming the coming era the age of institutional compression, defined not by who holds the largest workforce but by who can compress capability fastest without collapsing internally. This dispatch has run that thesis through one industry, end to end, on evidence. The fund after headcount is simply the first fully worked example of the institution after headcount, and the series continues from here.

All figures are computed across the 38-firm reconstruction, and the three multi-practice platforms are excluded from the medians as outliers. Figures are reported at sample level only and are not attributable to individual firms.

The following is a condensed extract from the full role-semantics dictionary that is held in the research appendix. The third column gives the reliable markers of a person's actual position, which the title alone never supplies.

The following roster sets out the institutional roles of the African fund, each with the exact title strings that were observed in the 38-firm sample. Titles are aggregated across the whole sample and are deliberately detached from any structure, firm, size class, or region. The same role frequently carries three or four different titles, and several title strings appear under more than one role, which is precisely the finding. Where a title string is known to occur at only one firm in the sample, it is marked (u); a small number of such strings are inherently distinctive, and they are presented here in aggregate only.

R1. Apex control (fund leadership, committee chair, LP or owner trust) Managing Partner · Co-Founder & Managing Partner · Founder & Managing Partner · Co-Managing Partner · Managing General Partner · Managing Director · Chief Executive Officer · Founder & Chairman · Founding Partner (executive usage)

R2. Senior investment authority (committee votes, deal leadership, boards) Partner · General Partner · Senior Partner · Investment Partner · Founding Partner (also used for non-executive originating sponsors: the same string denotes a different institution)

R2-lateral. Senior functional authority carried at partner grade Partner & Chief Operating Officer · ESG & Impact Partner (u) · Group CFO at partner grade · Operating Partner (full-time commercial usage)

R3. Delegated deal leadership (leads or co-leads transactions, holds board or observer seats, sits below the partner group) Principal · Investment Principal · Investment Director · Head of Investments · Senior Investment Officer · VP, Investments · Principal Investment Manager (u) A caution: Principal also appears in the sample as a pure grade applied to non-investment officers; see R9 and R12.

R3.5. Intermediate senior execution (between execution and deal leadership; ladders grow this rung organically) Senior Investment Manager · Associate Investment Director (u)

R4. Execution (sourcing support, diligence, models, memos, committee papers) Investment Manager · Senior Associate · Senior Investment Associate · Investment Associate · Associate · Vice President (execution-grade usage) · Transactor · Investment Executive · Senior Investment Executive

R5. Junior analytical execution (present at roughly one firm in four, with one or two heads) Analyst · Investment Analyst · Research Analyst A trap: Financial Analyst appears in the sample as a fund-finance seat, not a deal seat; see R10.

R6. Portfolio management (named ownership of post-investment relationships; increasingly a distinct third chain) Portfolio Manager · Director, Portfolio & Strategy (u) · Portfolio Associate · Value Creation lead · Operating Partner (portfolio-facing usage) · Operator-Investor (u; publicly self-described, see Part VI) · Chief Executive Operator-Investor (u; likewise)

R7. Platform and founder services (ecosystem, value-add programmes, founder support) Head of Platform · Head of Platform and Operations (combined with firm operations) · Head, Platform & Networks (carried at Principal grade) (u) · VP, Platform & Portfolio Operations (u) · Head of Incubation & Product Development (u) · Platform Associate · Associate, Platform & Networks

R8. Investor relations and fundraising support (the trust itself is partner-carried; these seats produce and coordinate) Chief of Staff (observed carrying investor relations, reporting and firm operations in one seat) · Senior Associate, Investor Relations & Communications · Executive Assistant / Investor Relations Associate (one combined seat) · Head of Investor Relations (large-firm usage)

R9. Fund finance ownership (signs the accounts; owns the administrator relationship, the audit and the valuations) Chief Financial Officer · Group CFO · Principal & Chief Financial Officer (u; the grade-not-role evidence) · Head of Finance and Fund Operations · Financial Controller · CFO/COO (one combined seat)

R10. Finance and fund-operations execution Financial Accountant · Accountant · Fund Operations Officer · Associate, Fund Operations · Financial Analyst (fund-finance usage) · Fund Administrator (in-house usage)

R11. Firm operations (running the manager itself: process, policy, premises, people logistics) Chief Operating Officer · Partner & COO · Director of Operations · Operations Director · Administrative Executive · Head of Platform and Operations (the operations half of the combined usage)

R12. Legal (in-house at control PE and at a minority of other firms; external counsel elsewhere) General Counsel · Principal & General Counsel (u; grade-not-role evidence) · Legal Counsel · Senior Associate, Legal

R13. Compliance and risk (in-house where the LP register demands it) Head of Compliance & AML (u) · Compliance Analyst · Chief Risk Officer (observed only at the largest firms)

R14. ESG and impact (the function is constant; the grade tracks the weight of DFI capital) ESG & Impact Partner (u) · VP, Sustainability & Impact (u) · Director of Social & Environmental Value (u) · Head of Sustainability & Impact · ESG Officer

R15. Talent and people (two different functions under one vocabulary: portfolio-facing talent support and internal HR) Head of Talent · Director, Talent Development · Manager, HR

R16. Communications and administration (the most frequently combined seats, and under compression the most fully subsumed) Communications Associate · Office & Marketing Manager (one combined seat) · Manager, Administration · Office Administrator · Executive Assistant

R17. External and fractional senior capacity (real authority that never sits on the core payroll) Venture Partner (advisory usage) · Venture Partner with a formal investment-committee seat · Operating Partner (fractional advisory usage) · Non-Executive Director · Non-Executive Chairman · Board Chair · Investment Committee Member · Advisor

Three observations become visible at a glance from this roster. First, the deal-leadership layer, which is R3, carries seven different title strings across the sample for what decision-rights analysis shows to be one institutional role. Second, four title strings in this roster appear under two different roles each, namely Founding Partner, Operating Partner, Financial Analyst, and Head of Platform and Operations, which means that the string alone can never locate the person. Third, the compound titles, meaning Partner & COO, Principal & General Counsel, Principal & CFO, CFO/COO, and the combined EA/IR seat, are not curiosities: they are the visible trace of function-combination, and they are the smallest observable units of the very compression that Part IV argues software now extends.

All web sources were accessed in June and July 2026. Sources are grouped by the role they play in the dispatch. Consistent with the anonymisation protocol, no source is listed for any anonymised firm: the fund-by-fund source register forms part of the proprietary research corpus described in the following section.

Series documents (Odit Frontier Partners).

• Odit Frontier Partners, AI Subsumption and the New-Age Disruptor, AI Subsumption Series Dispatch 01, Version 2.0, May 2026.
• Odit Frontier Partners, The Bypasser, AI Subsumption Note 04 and one-page companion AIS-04.1, July 2026.

Named case firms, quoted on their own public claims.

• Secha Capital: the firm's website and published descriptions of the Operator-Investor model, sechacapital.com.
• Kadan Capital: the firm's website, team pages and founding partner's published insights, kadan.vc, including kadan.vc/team and kadan.vc/in.
• Ripple Ventures: founder statements on the Ripple OS multi-agent system as reported by The Logic, thelogic.co, and the firm's public materials, rippleventures.com.
• Vela Partners: the firm's public self-description as an AI-native quantitative venture firm, vela.partners.
• Moonfire Ventures: founder and partner interviews on the firm's sourcing and evaluation systems, moonfire.com, and the profile at aiexpert.network.
• SignalFire: the firm's public description of the Beacon platform and externally reported fundraising, signalfire.com.

The adjacent AI-native services field.

• The 2026 market map of self-described AI-native service firms compiled by Nikola Lazarov, as covered by VC Cafe, vccafe.com.
• Anthropic, announcement of the AI-native enterprise services venture with Blackstone, Hellman & Friedman and Goldman Sachs Alternatives, anthropic.com/news.

The developed-market baseline.

• Publicly reported headcounts of the largest listed alternative-asset managers and conventional asset managers, per the most recent annual reports and investor pages of KKR (kkr.com), Blackstone (blackstone.com) and BlackRock (blackrock.com).

The 2026 AI tool landscape referenced in Part VI.

• n8n, Enterprise AI Agent Development Tools 2026, n8n.io/reports/2026-ai-agent-development-tools.
• n8n blog, We Need to Re-learn What AI Agent Development Tools Are in 2026, blog.n8n.io.
• StackOne, 120+ Agentic AI Tools Mapped Across 11 Categories, stackone.com/blog/ai-agent-tools-landscape-2026.
• Aishwarya Naresh Reganti, The AI Agent Stack in 2026, thenuancedperspective.substack.com.

Sample sources, by category. The 38-firm reconstruction draws on the firms' own team pages and published materials, professional biographies, deal announcements, promotion histories, regulatory disclosures, and appraisal and disclosure documents published by development finance institutions including the IFC, FMO, Norfund, Proparco, DEG, IFU and the European Investment Bank. The fund-by-fund register of these sources is held in the research appendix.

This publication is one visible output of a substantially larger proprietary research programme conducted by Odit Frontier Partners. The findings, frameworks and analysis presented here are anonymised by design; the underlying non-anonymised research corpus remains the exclusive intellectual property of Odit Frontier Partners.

That corpus includes, without limitation: non-anonymised organisational reconstructions of every firm in the sample; institutional architecture datasets; organogram reconstructions; canonical title mappings; decision-right mappings; governance mappings; reporting-line datasets; structural taxonomies; archetype classifications; institutional weight-class assignments; AI subsumability assessments; workflow and accountability classifications; confidence gradings; reconstruction metadata; evidence libraries; working papers; analytical notes; proprietary methodologies; and the prompts, reconstruction workflows and analytical tooling developed during the research programme. These assets constitute a proprietary institutional research corpus, and they are treated as such rather than as data incidental to a publication.

The research assets described above are not available for sale. They are not publicly accessible, and they are not licensed separately from advisory work. Portions of the research corpus may be used only within carefully scoped advisory or research engagements, where doing so is necessary to deliver the agreed work, and only to that extent.

Making an enquiry or requesting an engagement does not create any entitlement to access these research assets. Odit Frontier Partners retains sole and absolute discretion to determine whether any engagement will be accepted, and likewise retains sole discretion to determine whether any portion of the underlying research corpus will be used within an accepted engagement. Odit Frontier Partners reserves the right to decline any engagement or request for access, in whole or in part, without being required to provide reasons or justification.

This publication represents the analysis of Odit Frontier Partners. Readers are free to agree or disagree with its conclusions, and the firm assumes no obligation to persuade every reader. The publication is offered for analytical consideration rather than consensus.

Odit Frontier Partners assumes no obligation to disclose proprietary datasets, reconstruction files, working papers, confidence gradings, internal methodologies, analytical workflows, prompts, intermediate analytical artefacts, or any other proprietary research assets beyond what it has chosen to publish. Readers are invited to evaluate the publication on its own merits; nothing in it creates any entitlement to further disclosure.

Methodology and claims note. This dispatch draws on a structural reconstruction of 38 investment firms across African markets, built from public sources: team pages, professional biographies, deal announcements, promotion histories, regulatory disclosures, appraisal documents published by development finance institutions, and the firms' own published materials. Reporting relationships that could not be confirmed are treated as inference throughout the underlying research and are graded accordingly. All organogram examples in Parts II and III are syntheses recombined from multiple firms; no example reproduces any single firm's structure, and the mapping from examples to source material is held privately. Firms are named only where they have publicly self-described the models discussed (Secha Capital, Kadan Capital, Ripple Ventures, Vela Partners, Moonfire Ventures, SignalFire), and all operational figures attributed to those firms, including agent counts, throughput multiples, cost comparisons and headcount-equivalence claims, are the firms' own public statements: they are self-reported, unaudited, and quoted here as claims rather than as findings. The full fund-by-fund reconstruction, the complete role-semantics ontology, the compression scoring and the institutional matrices are held as a research appendix.