TTB 5: Modularity, Recombination & What Comes Next
How unbundling decides how a company can innovate, who gets to drive it, and which kind of institution it becomes.
In 1990, Rebecca Henderson and Kim Clark wrote a seminal paper that introduced a term describing a subtle yet often deadly threat to incumbents: architectural innovation. The idea of architectural innovation refers to reconfiguring how existing market components are combined and related: an overhaul approach that, as it often undermines the processes and routines companies have optimized over years, sits beyond the radar of incumbents.
Below, an unmissable episode of the Boundaryless Conversations Podcast where I spoke about architectural innovation - among other things - with Dave Gray
A recent 2024 study by Likun Cao, Ziwen Chen, and James Evans, titled “Modularity, Architectural Innovation, and New Venture Success,” offers, I believe, a good integration to the seminal view of Henderson and Clark. The paper looks at the innovation question by turning the “Architectural Innovation” lens on the entrant instead of the incumbent: for startups, they argue (with numbers) innovation often means assembling mature existing modules from distant economic sectors into a new configuration nobody has tried, and it does it by examining nearly 300,000 startups over the past 50 years.
That is the thread of this issue: unbundling - something I advocate for to organizations as fundamental - is not - and it shouldn’t be - just housekeeping. It decides whether an organization gets to innovate at all in an economy where innovation increasingly means rearchitecture rather than invention; it decides whether that innovation can be driven by the many rather than monopolized by the board, and as I will argue toward the end, it decides which kinds of institutions at all an organization is even able to become in an era that may require new institutional forms to be created.
Architecture beats invention
Cao, Chen, and Evans’ paper presents architectural innovation enabled by modularity as the safest and most successful strategy, while explaining how inventing new modules and incrementally swapping one module within an existing stack both increase the odds of failure. The magnitude is striking: across almost 300k U.S. venture-funded firms from 1976 to 2020, a maximal shift toward architectural innovation raises the likelihood of an IPO or a high-value acquisition, while a comparable shift toward modular innovation lowers it.
Uber is the canonical case; GPS, digital payments, two-sided marketplaces, reputation systems, and contractor labor were already on the shelf, and the innovation lived entirely in the rewiring.
What I keep seeing in the field with my (wisest) clients is the operationalization of this principle, and it has a shape we have described before. In TTB 4 I laid out the recurring three-layer pattern, core, supporting, generic, and the progressive modularization cycle (Innovate, Leverage, Componentize) that animates it, so I will not reinstate it fully here: the point that matters is to understand the dynamic.
Core units, close to the market, develop genuinely new capabilities; over time, those capabilities should be progressively modularized and pushed down into supporting units, often as modular technology platforms (node micro-enterprises in Rendanheyi, for example), where they harden into components that other units, first and outside partners later, can build on cheaply.
That is the supply side of architectural innovation: componentization is the act of stocking the shelf. Where the paper measured the outcome as a one-off founding bet happening in startups, what I see in composable organizations is a continuous operating property that becomes ever more important as we go: to innovate, your organization must first survive in a rapidly changing world.
This - as often happens - is actually not new: Herbert Simon provided the deep rationale for considering both perspectives (adaptation - and not only architectural reinvention) in his seminal 1962 paper “The Architecture of Complexity” through the concept of nearly decomposable systems. Modular assembly, he argued, hedges against catastrophic failure: near-decomposability (the property by which a complex system can be broken into subsystems that interact weakly across boundaries and strongly within them - modules) is not just a nice design preference but a key property of a system that wants to be adaptable.
Systems without it accumulate fragility faster than they accumulate capability. The organizational corollary is that capability nodes (or micro-enterprises) we work with at Boundaryless are not just fancy teams with unusual names but an attempt to deliberately engineer near-decomposability into an organization, so that each unit carries high internal coherence, end-to-end accountability, and a bounded surface of contractual exposure to its peers.
While explaining why that structure survives, Simon left open how autonomous modules coordinate once the hierarchy thins out: that is the gap where - I believe, today more than ever with the advent of AI - contracts, shared context engineering, shared platform layers, and (agentic) interfaces have to do the work.
From what we can do to what we value
There is a distinction the paper does not dwell on, but that I believe matters enormously. While modular invention and innovation (where we innovate a single new capability) enlarge what an organization can do and the stock of new things it can make, architectural innovation is a move in the space of outcomes.
From a given supply of capabilities, architectural innovation selects which configuration to actualize and, therefore, which result to pursue.
While capability building is close to value-neutral (the same modules wire into a dozen different ends), architecture is where the ends are chosen, and the moment you choose ends, you are expressing which values you intend to generate, which things you will and will not do, and which ecosystemic relationships you decide to feed.
You cannot recombine a monolith
Put the finding to work, and a single argument will follow: if architectural innovation is the winning move, and it is the recombination of modules across boundaries, then an organization’s capacity to innovate is bounded by two quantities: the supply of mature modules it can reach, and its own decomposability into modules that can take part in someone else’s recombination.
Inward, an organization that has unbundled itself, enabling platforms separated from vertical solutions, capabilities made discoverable and contractible, can recompose its own modules around a new opportunity without a reorganization and without a budget cycle, so the operating model becomes the execution environment rather than a slide about it.
A monolith cannot follow because its capabilities are fused, vertically integrated, its interfaces are implicit, and every recombination becomes a renegotiation of identity. Outward, in the direction the Cao paper implies, architectural innovation at the scale of an economy depends on a supply of modules that others can actually pick up; to be assembled into someone else’s winning configuration, or to fold theirs into yours.
Your organizational capabilities have to exist as modules with explicit interfaces, and if you are a monolith, you are not on the shelf, so the innovation proceeds around you with other people’s parts. Unbundling is how you innovate on yourself, and it is also how you earn the right to participate in innovation you did not originate.
Architectural innovation cannot be a board monopoly
Here is where I want to push past the conclusions in TTB 4: if architectural innovation is continuous rather than a founding bet, then someone has to keep driving it, and the default answer that almost every company I have worked with initially gives is that it falls to the board. Well, that is the failure mode hiding in plain sight.
When the center is the only actor allowed to run a recombination, the organization can run only as many architectural experiments as the board has management capability for, and the whole promise of modularity collapses back into a bottleneck. A unit that spots a way to combine its capability with a peer’s or an external partner’s has to queue for permission, and by the time permission arrives, the configuration may have moved.
The way out is to split the responsibility rather than centralize it. The board keeps two genuinely strategic acts: evaluating investments, which is to say allocating capital, and setting the enabling constraints the whole system runs inside. These are two genuinely important responsibilities that a board cannot shave off. In my experience, boards or founders often understand capital allocation but miss their responsibility to create reconfiguration and collaboration frameworks that enable units to recombine capabilities.
If every agreement between units, or between a unit and a partner, has to be improvised from scratch or escalated to the center for translation, distribution is a fiction. What units need is a common grammar of agreements (missing across almost every - even the latest - organizational-design schools): a way to strike, compose, and honor agreements peer to peer, so that architectural innovation becomes something the edges of the organization do continuously, not something the top signs off on occasionally.
That’s why we’ve created O2A and Orchestrion at Boundaryless: we believe these languages don’t have to be reinvented all the time.
Choosing the constraints is the instituting act
In his latest “Shaping Future Institutions” Geoff Mulgan (2026) - a British academic, author, and social entrepreneur, Professor of Collective Intelligence, Public Policy and Social Innovation at UCL who previously led Nesta and the Young Foundation - walks through economics, computer science, organizational studies, complexity theory and design, asking what each contributes to the design of institutions, and finds that none of them systematically studies how to design new ones: people building public institutions he argues still reach for options drafted half a century ago.
We hosted Mulgan on the Boundaryless Conversations Podcast a few years ago, and he was already arguing for an extended discipline of institutional architecture. In his latest, he argues that institutions are, in fact, information-processing architectures, and that:
“modular institutions with clearly defined interfaces between components are easier to reform, audit, and upgrade than monolithic ones”
We have written about the shift toward regenerative constraints: in a recent article, we argue that regenerative organization development (which effectively means institutional innovation) requires moving beyond just profit and loss as the sole forcing function. We propose extending decentralized, entrepreneurial operating models to multiple forms of value (financial, ecological, social, cultural, and learning capital). The challenge - we believe - and the failure mode of many attempts to institutional impact is not merely to measure this value (which is certainly a big deal), but to embed it into interfaces, incentives, budgets, and decision-making so that the innovation remains architectural, adaptive, and non-bureaucratic.
Mulgan’s 2026 essay is the same call, three years on, now accompanied by documentation showing that the discipline still has no home: a structural vacancy that someone has to fill with an operational rather than aspirational method.
Why this needs more than APIs
This is the vacancy Boundaryless has spent years formalizing in the O2A (Open Organization Alliance) standard, a grammar that lets every organizational node be at once a structural module, with clear boundaries and contracts, and a semantic module, able to declare its offerings, constraints, and dependencies in machine-readable form. At the moment, the O2A covers basic contract grammar, agreements, purchase, revenue split, and investment, plus composition- and outcome-based triggering, where a value you decide to generate becomes a milestone a contract can settle against; it is the base language the entrepreneurial units need to coordinate architectures among themselves. That is what turns “we are modular” from a diagram on a wall into something a partner, or an agent, can act on. The O2A will grow in time to include more and - we hope and believe - become one of the attempts to build architectural language for organizational and institutional innovation needed in the 21C, and we are searching for allies, so reach out to us if you want to be one.
It is tempting to file all of this - the regenerative, the societal, the institutional - under someone else’s problem. In Beyond Tomorrow (2026), the BCG Henderson Institute, not an activist think tank, names embracing a broader societal role (along with organizational modularity) as one of five low-regret moves that pay off across every scenario it modeled. As public institutions fray, companies are being drafted to carry workers’ well-being, local resilience, crisis management, and the needs of the communities they operate in. That quietly dissolves the lazy distinction we lean on when we think the institutional problem is out of reach for commercial organizations.
Institution is just a durable set of constraints that allocates value over time, and that is what an organization becomes the moment it takes its constraints seriously. The membrane between the two is thin, and thinning. Choosing the constraints was always institutional design; we were just calling it strategy.
Below, a recent episode of the Boundaryless Conversations Podcast where I spoke with Nikolaus Lang about Beyond Tomorrow from BCG Henderson Institute
The times in which firms competed on what they owned are transforming into one in which they focus on what they can mobilize and contribute. Cao, Chen, and Evans highlight this shift from the perspective of newcomers succeeding by recombining existing modules rather than inventing new ones; Mulgan honestly admits that the discipline for designing institutional innovation is lacking; and our contribution is the operational layer that supports both ideas.
Curated Links
The Coming Organizational Meltdown
Tobaccowala’s fusion of IT and HR and the death of the job as an atomic unit is the clearest recent statement of superstructure dissolving: the container melts before anyone builds what it was implicitly holding.
AI Agents as Coordination Technology
Kosters frames agents as the infrastructure that collapses coordination costs, making composition viable at scale; the caveat is that the collapse is uneven across codifiable and contextual knowledge.
Post-agent companies
The “give away the labor, capture the network” logic is a live instance of the O2A wager: when cognitive labor approaches zero, the defensible layer is the contractual and semantic one (who does what, for whom, under which agreement) > context engineering becomes ecosystem engineering.
Shaping future institutions
Mulgan diagnoses the absence of any institutional design science, and one line does the work of a manifesto: “Modular institutions with clearly defined interfaces between components are easier to reform, audit, and upgrade than monolithic ones”; in short, our thesis.
Modularity, Architectural Innovation, and New Venture Success
Cao, Chen, and Evans provide the empirical backbone: across 300,000 ventures over 45 years, recombining mature modules beats inventing new ones; the evidence that composition, not creation from scratch, is the economically viable path.
The Rebel Alliance
USV maps the agentic AI stack as structurally distributed, forcing open interfaces and layer specialization.
Work Updates
A dense fortnight, with two releases we’ve been building toward for a long time.
O2A is live. We shipped the first version of the O2A (Open Organization Alliance) standard alongside the new Boundaryless website. This is the thin-waist layer the essay above keeps pointing at: the minimal canonical model of nodes, contracts, offerings, and ledger events that makes capabilities composable for humans and agents. It is now public and inspectable.
Orchestrion is open for private demos. Our stateful execution OS, which implements O2A, is available in private beta for our customers. This is where the model stops being a diagram and becomes a live, auditable composition runtime: would you like to understand your gaps to be ready to use it? Just reach out.
All Boundaryless methodologies are now searchable. Years of methodology that lived inside monolithic, non-indexable documents are now published on our documentation website alongside O2A. Someone called it “Alexandria’s library.” It is far more useful as a queryable resource than it ever was as a set of PDFs and, fittingly, it makes our own context legible.
Platform strategy in the field. We are running a strategy engagement with a client moving from being one player in their ecosystem to becoming the ecosystem organizer. The challenge is the durable one: how do you turn competitors into customers, and help every professional in the space become the best version of themselves?
If your organization is feeling the meltdown Palmer describes (management thinning out, capabilities that exist somewhere but cannot find each other…) I’d genuinely like to hear about it.