Why the race to make your company machine-readable is the fragility risk of the next three years
A particular meeting is happening in boardrooms right now, and I have sat on both sides of the table. An adviser explains that the enterprise must become legible to artificial intelligence: the processes, decision rights, and tacit rules currently living in people’s heads have to be written down, structured, versioned, and made executable. The room nods. Someone says “single source of truth”. A budget line appears, and with it a three-year programme.
The logic is close to unimpeachable. An agent cannot act on what it cannot read, and most enterprises are illegible to their own systems, let alone to a model. Boston Consulting Group has given the idea the cleanest formulation in its “Enterprise as Code” thesis: explicitly define how the business operates so both people and AI agents can understand it, test it, and improve it. BCG’s analogy is apt; just as cloud computing democratised access to infrastructure, codified operations democratise access to operational capability.
I think the diagnosis is right and the prescription is dangerous. My contention is this: the enterprises that codify hardest over the next three years will not become the most adaptable. A meaningful number of them will become the most brittle organisations in their sector, and they will not discover it until something arrives from outside the specification.
What the Prussians learned about legibility
The anthropologist James C. Scott, in Seeing Like a State (Yale University Press, 1998), described how states seeking to manage forests, cities and populations first had to make them legible, standardised, mapped and counted. His founding example is Prussian and Saxon scientific forestry in the late eighteenth century. Foresters replaced the chaotic old-growth woodland with the Normalbaum: a single commercially optimal species, planted in rows, of uniform age, countable from a desk in Berlin. Yields in the first rotation were superb. The second rotation collapsed, and Scott borrows a German phrase for this: Waldsterben, which literally means forest death.
The specification stripped out everything it could not price: the undergrowth, the fungal networks, the insects, the deadfall, the apparently unproductive mess that had, in fact, been running the nutrient cycle. The forest that could finally be counted turned out to be the forest that could no longer sustain itself.
That is the risk sitting inside every enterprise codification programme, and it is not a risk the vendors will raise with you. A specification captures what the organisation does. It cannot capture the substrate: the account director who knows a client is about to defect because of a change in tone on a call; the operations manager who quietly ignores the approved supplier list during a load-shedding week because she knows who will actually answer the phone.
In the language of Nassim Nicholas Taleb’s Antifragile (Random House, 2012), that substrate is where an organisation’s optionality lives. Attempt to code it, and you do not capture it. You evict it because the moment a judgement becomes a documented rule, the person exercising it stops being paid to think and starts being audited for compliance.
Codify the eighty. Defend the twenty.
The instruction is not “do not codify”. That would be a counsel of nostalgia, and nostalgia is not a strategy. The instruction is that codification must be a deliberate, budgeted and defended split rather than a programme that runs until it hits the edges of the organisation.
Code the eighty per cent that is repeatable, high-volume and low-variance: billing, onboarding, reconciliation, compliance reporting, media trafficking, campaign quality assurance. Do it aggressively, and do it faster than you currently plan to, because that work is where the agentic productivity gains are real and where your competitors will find them within eighteen months.
Then ring-fence the twenty per cent that must not be written down: entry into an unfamiliar market, pricing under genuine ambiguity, creative judgement, crisis response, and the moment a client’s own business model changes shape. These are domains where the correct answer is not yet a pattern, and where encoding last year’s pattern is materially worse than having no pattern at all, because it is confidently wrong at machine speed.
The test I now use is crude and effective. If you can write the rule without using the word “depends”, code it. If the honest answer begins “it depends on…”, it belongs to a named human being with a real mandate.
The discipline most enterprises will fail is the second half. Codified work is measurable, and measurable work wins budget arguments against unmeasured work every single time. Without an explicit defender, the coded eighty will colonise the twenty by ordinary organisational gravity, one reasonable-sounding process improvement at a time. Somebody senior has to hold the line, and it should be a standing role, not a sentiment.
Agility is not adaptability
This distinction matters more than the industry’s loose usage suggests. McKinsey & Company frames adaptability as a proactive meta-skill of continuous learning, adjusting ahead of change, while resilience is the capacity to recover after it. Agility is neither. Agility is the muscle: the speed at which you execute a shift once the decision has been taken.
Most large enterprises have spent a decade buying agility and calling it adaptability. They have quarterly planning cycles, squads, tribes and stand-ups, and they are now very fast. Speed without adaptability is simply an efficient way to keep doing the wrong thing for longer. Enterprise as Code will make you faster still. It will not, on its own, make you wiser, because wisdom is a property of the twenty per cent.
Decision-making is the bottleneck, and consensus is the clog
If adaptability lives in the uncoded twenty per cent, then decision-making is its rate limiter. McKinsey’s work on organisational decision-making identifies the trap precisely: a one-size-fits-all approval path in which an ad-hoc call, a big bet, a cross-cutting decision, and a delegated decision are all routed through the same committee at the same tempo. The result is that the trivial is over-governed and the consequential is under-examined.
Underneath that sits the deeper problem of false consensus. Pursuing unanimity does not produce alignment; it produces a diluted decision that nobody owns and a room full of silent dissenters who will be proved right in eight months. Agile practice offers a blunt corrective in “fist-to-five,” asking every participant to show, on their hand, how strongly they actually support a decision, which converts polite silence into data in about four seconds.
This is where psychological safety stops being a wellbeing initiative and becomes infrastructure. Amy Edmondson’s original research (Administrative Science Quarterly, 1999) defined it as the shared belief that the team is safe for interpersonal risk-taking. In a codified enterprise, its function becomes considerably more concrete: the specification will be wrong at some point, and the only early-warning system you have is a person junior enough to see the error and safe enough to say so before it is executed ten thousand times by an agent that does not doubt itself. Psychological safety is not the soft counterweight to the code. It is the error-correction layer.
The human-centred build
Deloitte’s framing of the adaptable organisation puts four pillars around this: build on purpose and meaning; organise for effectiveness by unravelling complexity and simplifying the corporate centre; optimise for a future workforce that increasingly sits in the ecosystem rather than on the payroll; and design for human-centred realities through inclusive design and genuine empowerment. Read alongside the argument above, the second pillar is the operative one. A simplified corporate centre is not an efficiency measure. It is what leaves room for judgement to survive at the edge, where the signal actually arrives.
An emerging-market advantage
I run a business on a continent where the specification has never been stable. Currency moves, grid failures, regulatory turns and supply chains that vary by week have meant that African operators long ago built the capability that global enterprises are now trying to procure: the reflex to hold a plan loosely and a purpose tightly. That is not a romantic claim about improvisation. It is an observation that permanent volatility is an expensive but effective training regime. Firms that have learned this have a genuinely transferable asset as volatility becomes the global default.
What are you able to do immediately?
Three things, all of which fit inside a quarter.
- First, take your codification backlog and colour it: what proportion is honestly low-variance, and what proportion is judgement wearing a process costume? If the second number is above a fifth of the programme, you are building a monoculture.
- Second, name an accountable owner for the uncoded twenty per cent, with the standing and the budget to refuse codification requests.
- Third, take your slowest high-stakes decision, classify it properly, assign a single accountable decision-maker rather than a committee, and run fist-to-five on the next call. You will learn more about your organisation in that one meeting than in the last two engagement surveys.
Anti-fragile enterprises are not the ones with the best specification. They are the ones that know precisely which parts of themselves must never be specified, and have the courage to defend that ground when the efficiency case says otherwise.
That is the brave enterprise, and building one is a leadership decision long before it is a technology decision.
