The (AI) Gardener Also Builds the Beds

Within the past week I published two pieces here that, read side by side, could appea to have two slightly opposing views included. Here’s the thing sitting underneath both and why it’s the most useful thing I can tell you about doing this job for real.

Someone I really respect, quite reasonably, noticed.

The first piece argued that AI adoption is not a rollout. It’s an ecology. Uneven, social, impossible to install, and the job isn’t to deploy a tool but to tend the conditions, the way a gardener does. Stop thinking like an engineer, I said. You can mandate access. You cannot mandate that people get good at thinking with the thing.

The second laid out an AI-native product operating system in seven proposed layers. Signal sensed continuously, decisions traceable to commercial logic, feedback loops that close on a trigger. The kind of thing you could draw on a whiteboard and audit on a Thursday. It feels unapologetically like my old process improvement days and engineering logic abound.

So one of two things is true. Either I changed my mind somewhere between the two posts, or the reality of this job means there is something underneath both that makes them the same argument rather than opposite ones which they could appear on the surface. It’s the second obviously. And the seam where they meet is, I think, the whole job and the part that’s hard. Really hard.

The contradiction, said plainly

It’s worth being honest about how sharp the tension looks before pretending it away, because if I gloss it you won’t trust what it is really like to do this work.

The ecology piece feels anti-engineering at its core. Its whole point is that capability is a living thing — it spreads through people, socially, on a timetable nobody controls, and the managerial instinct to standardise it is precisely what kills it. Treat the organisation as a machine, define the component, install it, confirm it’s running — and you’ve made a category error about what kind of change you’re dealing with.

The OS piece is a machine. Seven proposed layers (for simplicity of an article), each wired to the next, the explicit promise that the whole thing should keep running after any individual leaves the room. That is a blueprint. It is the thing the first piece told you to put down potentially.

Put them on the same table and one of them appears to be wrong.

Neither is. They’re describing two different layers of the same reality, and the best people I know can do both, make it clear when they are doing so and find the right balance to walk both lines.

The same villain, twice

Start with what each piece is actually frightened of, because it turns out to be the same thing wearing two outfits.

The ecology piece is frightened of the discovery that dies in one person’s head. Someone three floors down works out exactly how to do a task brilliantly, tells nobody, and the chain that should have carried that capability to the next desk snaps at the first link. Value, trapped where it landed.

The OS piece is frightened of the inisght that dies in a Notion doc. A customer insight gets summarised, looks like progress, and then sits there until the next quarterly planning cycle while the thing it should have changed goes unchanged. Value, trapped where it landed.

Those are the same villain at two altitudes. Desk-to-desk, and insight-to-P&L. The ecology piece is about capability that can’t travel between people. The OS piece is about intelligence that can’t travel between functions.

In both, themost important problemm and the failure mode is identical: something true and useful gets discovered and then has nowhere to go.

T two pieces aren’t rival theories of AI which they could seem like. They’re the same complaint:

Value gets stuck in AI Adoption

Attempting to be told two days, once at the level of the human and once at the level of the system.

Tend the people, engineer the flow

Here’s the reframe I’d actually put forward.

The mistake nearly every company makes is not that it engineers too much or tends too little. It’s that it engineers the wrong half and tends the wrong half. It does both, just in the wrong places.

It engineers the people:

Mandates
Approved tools
Training booked
Leaderboard of logins
Number on a slide

Machine thinking applied to human capability, which is exactly the error the ecology piece is about. You cannot install someone becoming good at reasoning with these tools, and every attempt to do it by mandate produces the 87%-adoption slide sitting on top of work that looks identical to last year’s.

The other error is that the company leaves the flow to chance:

Whether a discovery reaches the next person
Whether an insight reaches a revenue conversation
Whether last quarter’s bet gets checked against the metric it was meant to move

All of that is left to goodwill, good memory, and whoever happens to be in the room. Which is exactly the error the OS piece is about.

The right thing to do in my view, and the experienced operator move is to flip both.

Tend the people. Patiently, socially, with judgment, on a timetable you don’t fully control. That half (or maybe more than half) is a garden and it always will be.
Engineer the flow. ruthlessly, deliberately, with as much structure as you can stand. That half is plumbing, and it should be. The path that carries what one person learned to the next person, and from there into a commercial decision, is not something you tend and hope for. It’s something you build.

Tend the people. Engineer the flow. Most companies do exactly the reverse: they engineer the people and leave the flow to chance. Then wonder why their best ideas keep dying three desks from where they were born.

Fifteen minutes to change something

If you want the two pieces collapsed into a single object you can actually see, it’s this.

In the ecology piece I described the cheapest tending move I know: fifteen minutes bolted onto the front of a meeting that already exists, where one person shows the single most useful thing they did in the last fortnight, live, on their own screen, and takes two questions. No deck. No programme with a name. Just a short focus on real learning.

In the OS piece, the seventh layer is the operating rhythm, the cadence that keeps everything else running, owned, and honest.

Those are the same fifteen minutes. The gardener calls it tending. The systems person calls it a feedback loop closing. It is one instrument, and it happens to be simultaneously the most human thing in the building. That’s not a coincidence. It’s the whole point. The places where the ecology and the operating system are genuinely the same thing are the rhythms i.e. the recurring, owned moments where living capability gets carried into the system on purpose. Build those, and you don’t have to choose between the garden and the machine. The rhythm is both.

What it means to operate this way

So what changes on Monday, if you take both pieces seriously at once.

You stop trying to control the outcomes and start shaping the channels. You can’t dictate what your most capable people discover, that’s the garden, and Cynefin is right that you steer it by managing constraints rather than chasing targets. But you can build the beds: the recurring slots, the connections between product and commercial data, the loop that takes a signal and routes it to the conversation it should trigger. Beds don’t tell the plants what to be. They decide whether what grows reaches the kitchen or rots in the ground.

You keep the judgment human. The proposed seven layers will carry a decision faithfully from strategy to sprint (do people still Sprint now “Agile doesn’t work” ;) ?) without losing the reason it was made. What they will never do is make the decision for you, or tell you which experiment is quietly going somewhere and which confident, well-spoken one is a dead end. “Have a view” was the fourth tending move for a reason. The system is the circulation. You’re still the one deciding what’s worth circulating.

And you treat the structure as the thing that makes tending compound instead of evaporate. This is the part the ecology piece can’t do on its own. A beautifully tended garden with no beds, no paths, no irrigation still loses most of what it grows — the person who cracked something shows one colleague, and then where does it go? An operating system with no living capability flowing through it is worse: seven immaculately wired layers and nothing real moving along them, which is just the AI-washed failure with better diagrams. You need both. The ecology is where the value is created. The OS is what stops it dying on the way to anywhere useful.

The job, again

Strip both pieces back and they turn out to be one obsession seen through two windows.

The gardener builds the conditions so that people flourish without needing the gardener.
The operating system keeps running after any individual leaves the room.

Those aren’t two different ambitions. They’re the same one:

Capability that outlasts the person who built it.

It’s the thing I’ve spent most of my working life pointing at, whether the title on the engagement or job said product leader, consultant, fractional or anything else. The test of whether we did anything that mattered is what’s still standing, and still growing, once we’ve gone.

So the job hasn’t changed since the first piece. It’s still mostly walking the floor and noticing, finding the person a few rows down who’s doing things this year that weren’t possible last year and whom nobody senior has clocked yet. That’s still where I’d start.

I’d just add the second half now, the half I left implicit. Once you’ve found them, don’t only tend them. Build the bed they grow into and the path their discovery travels down, so that the next person, and the function carrying business targets, and the metric on the board’s slide all feel it too.

Tend the people. Engineer the flow.

The (AI) gardener, it turns out, also builds the beds.