VIMARSHA · FIELD NOTE 01 · AI STRATEGY · 5 MIN

Everyone has a map for your AI journey. That’s the problem.

THE QUESTION THIS NOTE SITS WITH
“We've changed our AI strategy three times in the last year. Things are moving faster than we can keep up.”

A founder said that to us recently, and there was no embarrassment in it — just fatigue. Most companies are sitting with some version of the same sentence. The goalposts have moved three times in twelve months, and every serious voice in the market seems to be pointing somewhere different.

Just this week, three answers crossed our desk. Each from a credible source. Each internally coherent. Each pointing a different way.

MAP 1 — AGENT-FIRST

The frontier lab's view: the individual becomes an organization

Anthropic's Steps of AI Adoption (Boris Cherny, July 2026) charts the climb from one engineer with one agent to one engineer steering a thousand. Read the roles he assigns the human at each step: a pair. An orchestrator. A manager of managers — an org tree. A VP, steering by intent.

It's a sharp map. And its direction is clear: maturity means the individual growing into an organization of agents. Start at the desk, scale up.

MAP 2 — ENTERPRISE-FIRST

DevRev's view: the substrate at the centre

At the DevRev Leadership Circle in Bengaluru this July, co-founder Manoj Agarwal laid out the opposite starting point: five layers, with the enterprise's organized memory at the core — the internet's memory and the individual's alongside, agents acting on all three, humans shaping the agents above. The individual and the organization circle the substrate.

Also sharp. Direction: maturity means the enterprise getting its context organized first — because agents without organized memory are noise. Start at the centre, build out.

MAP 3 — PAIR-FIRST

The third view: people + AI at the centre

The third school puts neither the agent nor the substrate at the centre, but the pairing — human and AI as complements, a yin-yang, each covering what the other can't. Maturity here means deepening that pairing across the workforce: every associate augmented, judgment kept human, capability grown together.

Also credible. Direction: start with people, mature the partnership.

THE ACTUAL PAIN

Three good maps. No one path.

Now stand inside a real organization and look around.

Your engineers are at three different steps of the agent-first climb — some pairing, some orchestrating, one quietly running an org tree. Your enterprise memory is half-organized at best. Some functions are deep in the pairing; others haven't started. Three vendor decks on the table, each drawn on a different map.

People sit at different stages. The maps point different ways. There is no one path — and that, precisely, is the entropy every AI program is drowning in.

This is the real state of the market's middle: not resistance, not ignorance — too many credible directions at once. Boards asking "where are we on the curve?" without agreeing which curve. Budgets split three ways to hedge. Committees formed to pick a map.

THE TURN

The answer is not one of the three

Here's the step back that dissolves the entropy: all three maps share one assumption — that AI uptake is the journey. Pick a starting point, climb, and maturity is how far you've climbed.

But uptake was never the journey. It's the vehicle.

The questions that actually order the work are older and a lot more basic: What are we solving for? Which outcomes? Which decisions drive those outcomes — and where do they sit today, on what context, at what speed? What does our operating model need so those decisions get made better, faster, closer to the customer?

Don't start with AI uptake. Step back to what requires solving — and how you're solving it. The maps become useful the moment they stop being the strategy.

Answer those questions and the three maps snap into place as a supply catalogue, not a direction. A decision starved of leverage? Pull from the agent-first climb. Decisions starved of context? That's the substrate map, and it's the priority. Judgment that must stay human while everything speeds up? That's the pairing, deliberately designed. You'll use all three — in the mix and sequence your decision anatomy demands, not the one on anyone's slide.

That's also the honest route through transition to transformation: a transition scoped by "what requires solving" keeps transformation reachable. A transition scoped by "how much AI have we adopted" matures the vehicle and forgets the journey.

Your operating model — the outcomes you're going after, and the decisions that fuel them — decides your AI strategy. Not your use-case list. And not, it turns out, anyone's maturity map either.

THE MAP TEST

Run this with your leadership team — it takes one meeting and usually ends it:

1
Ask each leader to name, privately, the maturity model the company is following. The number of different answers is your entropy — measured, not felt.
2
Capture the outcome the company — or the function — is solving for, and the two or three decisions that most determine it. If the answers don't converge, no map can save the program — the disagreement is upstream of AI entirely.
3
Take your single largest AI commitment — money or people. Trace it: does it serve one of those decisions, or a stage on someone else's map? Fund the answer accordingly.
Drowning in maps, short on a path?
Send one workflow — or one decision that bothers you. A one-page read comes back in 48 hours.
RUN ONE WORKFLOW THROUGH THE MODEL
SHARE  LINKEDIN  ·  COPY LINK  ·  X
Was this useful? Leave a comment
Vimarsha — field notes from the decision layer
One note a week. Real decisions, real workflows. No filler.