ORGANISATIONAL SPECIATION

How 2026 produces divergent forms

Our previous chapter argued that cheap intelligence reveals the organisational phenotype. The visible behaviour. The muscle memory. The habits that usually stay half-hidden behind the sheer effort of getting work done.

This final chapter goes one step further.

Because once you accept phenotype exists, something else must too - the process of selection.

And when selection runs for long enough, divergence follows.

In biology, speciation is what happens when a changing environment makes different traits advantageous in different contexts. Populations adapt. Over time, they split. Eventually, they become meaningfully distinct. They can still share an ancestor, but they stop being easily interchangeable.

Which means ‘speciation' is a strong word for organisations. It needs handling with care. Companies do not reproduce like organisms. They inherit traits genetically only loosely.

And yet.

In a cheap-intelligence environment, we should still expect two organisational ‘species’ to become more common, more stable and more different from each other than most leaders currently assume.

Because the environment is changing faster than the institutions inside it.

And because copying, in practice, is far harder than it looks on PowerPoint.

Why divergence emerges in 2026

For three years, AI has been a capability story. Demos. Benchmarks. Adoption claims. Internal enthusiasm that often felt real in the moment.

2026 is when it becomes an accounting story.

The Productivity Clock was always heading here. Not out of malice. Out of cadence.

By the middle of this year, many large organisations will have had close to two years of serious AI programmes: funded, staffed, governed, measured. Two years is enough time for pilots to harden into capability. It is also enough time for programmes to collapse into an expensive form of organisational theatre.

Because forecasts, budgets and operating metrics are where belief goes to be weighed. CFOs and boards do not argue with demos. They argue with confidence levels. They argue with auditability. They argue with whether a benefit survives scrutiny long enough to become guidance.

And that is where selection pressure enters.

Quietly. Through:

• Finance committees that ask whether the gains can be forecast.

• Risk functions that ask who owns the output.

• Audit that asks whether the process can be repeated.

• Customers who test your escalation paths at 4.55pm on a Friday.

• Regulators and litigators who only show up after something goes wrong.

  
  

The environment does not reward intelligence in isolation. It rewards intelligence that can be defended, repeated and owned. Which brings us to the split.

1. The high-trust, high-governance species

Regulated, audited, cautious and increasingly capable, these organisations thrive where the cost of failure is real and legible. Where mistakes are expensive in money, reputation, licence or lives. Finance, healthcare, infrastructure, defence, insurance, energy, pharmaceuticals. Also any enterprise where the centre of gravity sits in procurement, compliance and systems of record.

The defining trait of the species is ‘proof machinery’.

This is the organisation that treats AI like infrastructure. It assumes that scaling requires:

• Clear ownership

• Explicit evaluation criteria

• Version control and audit trails

• Monitoring for drift

• Structured exception handling

• Escalation paths that work under stress

• Human review that is designed, not improvised

• A definition of ‘good enough’ that can survive an uncomfortable meeting

It builds the boring parts first, or at least early enough that the boring parts are ready when the first incident arrives. It invests in legibility without suffocating the frontline. It learns to separate reversible decisions from irreversible ones, and it bakes that distinction into workflows.

The trade is obvious. This species moves slower at the surface. It ships fewer dramatic announcements. It sometimes looks timid beside a start-up’s velocity.

But it scales.

And it clears gates.

Most importantly, it creates the conditions under which AI productivity can show up where it has to show up: in budgets, forecasts and audited outcomes.

In this species, speed is accepted as abundant. Reliability becomes the currency. Trust becomes the compounding asset.

This is also the species that internalises the logic of high reliability research, even if it never uses that language. Preoccupation with failure. Sensitivity to operational results. Deference to expertise. A refusal to simplify away edge cases just because they are annoying.

It has mastered the organisational equivalent of building muscle around weak joints. Less showy, more survivable.

2) The high-velocity, low-constraint species

Shipping constantly, tolerating error and thriving where mistakes are survivable, this species thrives where the cost of failure is low enough to be metabolised. Consumer apps. Marketing. Sales tooling. Media. Many forms of software. Start-ups. Product teams inside larger organisations that sit far enough from the regulated core to behave like their own ecosystem.

Their defining trait is throughput.

This organisation treats AI as a force multiplier for iteration. It assumes that learning comes from shipping. It expects error, because error is information. It builds disposable tools on purpose. It replaces long decision cycles with short ones. It does not wait for certainty because certainty is expensive and often fictional.

Its strengths are the mirror image of the first species:

• Fast feedback loops

• Rapid prototyping

• Continuous deployment

• High tolerance for mess

• High willingness to discard work

• Local autonomy over central permission

• A cultural comfort with provisionality

  
  

It can move like a swarm. Make many small bets, have few committee gates. It uses cheap intelligence to compress the distance between idea and artefact.

This species accumulates risk differently. It does not naturally produce audit trails. It does not automatically produce legibility. It sometimes confuses speed with coherence. It can generate outputs faster than it can absorb them. Review debt becomes a permanent background hum.

And yet it can still be wildly successful.

Because in its niche, the market selects for learning speed, distribution and constant improvement. Customers reward novelty and convenience. Competitors punish hesitation.

This is the species that turns cheap intelligence into constant motion.

Why both species can win

At this point, most strategy discussions make a quiet mistake.

They assume that a successful organisation can simply adopt the other species’ traits when required. They imagine governance as a module you can add, or speed as a button you can press.

But real life is stickier.

Traits come in bundles. And those bundles reinforce themselves.

High-governance organisations tend to select people who are comfortable with scrutiny, documentation and shared accountability. They build systems where escalation is safe, responsibility is explicit and exceptions have a home. Over time, they become better at being reliable, because reliability is rewarded internally.

High-velocity organisations tend to select people who are comfortable with ambiguity, rapid iteration and local decision rights. They build systems where initiative is rewarded, shipping is status and the fastest learner wins. Over time, they become better at being fast, because speed is rewarded internally.

Each species becomes what it measures. Each avoids what it punishes.

Copying the other species is difficult for reasons that are mundane, not mystical:

• Incentives do not change cleanly

• Governance rewires status

• Velocity rewires authority

• Tooling embeds assumptions about ownership

• Hiring embeds assumptions about risk

• Budgeting embeds assumptions about proof

An organisation can mimic the surface rituals of the other species. It can buy the same software. It can borrow the same vocabulary. It can hire a few people with the right job titles.

But the phenotype will give it away.

Because phenotype is how the work actually happens when the meeting ends.

The failed hybrids, where selection gets brutal

A cheap-intelligence environment does not only produce two viable forms. It also exposes the unstable ones.

There is a particularly common failure mode we should expect to see more clearly in 2026: organisations that combine high constraint with low trust.

These are businesses with many gates and weak judgement. Lots of process, little clarity. Lots of review, little ownership. They build controls that nobody believes in, then they wonder why people route around them.

Cheap intelligence makes this failure mode louder.

Outputs multiply. Queues grow. Review debt compounds. The organisation starts to feel like a restaurant that can cook infinite meals instantly, but only has one waiter and a manager who insists on approving every plate.

A second failure mode we can expect to see is high velocity applied to high-stakes domains without the matching proof machinery.

This is where incidents happen. Not necessarily dramatic ones at first. Small errors. Small exceptions. Small misunderstandings that compound. Then a customer complaint. Then an investigation. Then a new control. Then a slow-down that arrives like winter.

Selection pressure does not negotiate here. It simply applies consequences.

What we expect to see by mid-2026

We do not expect a series of headlines that prove our hypotheses. We expect signals. Boring ones. The kind that show up in spreadsheets, budgets and hiring plans.

But here’s what they look like in practice:

Budget posture changes Programmes that cannot produce defensible proof get squeezed. Programmes that can show stable benefit get extended, scaled and normalised.

The language of proof Organisations talk less about adoption and more about evaluation, ownership and repeatability. They stop counting seats and start counting outcomes.

Procurement hardening AI moves from discretionary spend to governed spend. The organisation begins consolidating tools. Shadow usage is either brought into policy or pushed out.

Operating metrics bend, or they do not Margins widen with confidence, or they stay flat while activity increases. Headcount plans change, or they stay stubbornly similar. Forecasts absorb gains, or they refuse.

Exception handling becomes the tell High-performing organisations will not have fewer exceptions. They will have cleaner ways of catching them, routing them and learning from them.

Two-speed designs become more explicit The successful organisations will formalise a path from disposable tools to hardened systems. They will give experimentation a home, and they will give infrastructure a spine.

This is how selection will show up.

Because boards reward what clears scrutiny. Teams reward what ships. Labour markets reward what makes judgement teachable. Vendors reward workflows that map onto product. Regulators reward explainability after failure. Capital markets reward narratives that can survive contact with numbers.

Each of these is a selection force.

Together, they push organisations towards forms that can live in this environment.

The final reveal about 2026

So what will 2026 be, really?

Our bet remains simple.

2026 is when AI stops being the main variable. Organisational form will assume that role. Although most will be oblivious to it.

Model capability will keep rising. Tooling will keep improving. Output will keep getting cheaper.

The scarce resources will sit elsewhere:

• Judgement that can be trusted

• Review that keeps up with throughput

• Escalation paths that work when the system is stressed

• Ownership that survives a bad day

• Proof that clears forecast scrutiny

That is the heart of our thesis.

The Productivity Clock is the moment when belief meets budgeting. It is the point at which AI moves from interesting to accountable.

And once that accountability arrives, selection begins to bite.

Some organisations will look, from the outside, as if they have found a new gear. Others will look as if they are moving, yet going nowhere. Both may be using the same models.

Different phenotype. Different outcome.

Different outcomes, repeated for long enough, becomes divergence.

The environment is busy producing new viable forms.

And in 2026, as the clock starts ringing, we’ll start to see which forms can survive the new, some might believe testing, environment.