One method. Two moments in the ownership cycle.
We take a business that runs on one person, on data spread across tools that do not talk to each other, and on processes only that person can explain — and turn it into something that runs without them, and that an adviser can inspect. Whether you are preparing to sell or you have just bought, the work is the same. Only the starting point and the urgency differ.
Two starting points, one method.
We find the constraint before we propose anything.
We start with a discovery call — not a sales pitch, a genuine attempt to understand your business. Everything after it depends on getting the first question right.
- A discovery callHow the business actually runs, who holds what in their head, and what it costs when the person holding it is unavailable. No pitch, and nothing to sign.
- Find the real constraintWe look for the single bottleneck actually limiting the business — the one thing that, if fixed, moves every number that matters. (This follows Theory of Constraints, the same discovery method now used by private equity firms applying it across portfolio operations.)
- Test whether an agent is the answerOnly then do we explore whether an AI agent is the right way to address it. Sometimes it is. Sometimes the constraint is not something an agent can touch, and we will tell you that too.
- Scope one stage, not threeYou see the first stage scoped and priced on its own terms before anything else is discussed. Nobody commits to a programme they have not seen the shape of.
A diagnosis that ends in a slide deck and a sales pitch is not a diagnosis. If the constraint is not something we can help with, that is what you will hear.
The three stages, in detail
- Stage 1 — Stop the subscription sprawlNot a big-bang migration. The new custom app becomes where the work happens and writes into one central store from day one, and each legacy tool is replaced or piped in as its turn comes — so the business never stops running to accommodate the project. The point is not a smaller subscription bill. It is that the work stops living in four places only one person can reconcile, and starts living somewhere a new operator, or a buyer's adviser, can be shown.What you end up with
- A custom app your team actually works in
- One central data store, written to from day one
- A cutover order for the tools still in play
- A process a new hire can be handed without you in the room
- Stage 2 — Data as the foundationThe unglamorous 80%, and the stage a diligence process actually tests. Reconciling what four systems each believe to be true, resolving it into one structured record, and building business data that is clean, documented and trustworthy. A buyer's adviser will not ask whether you use AI. They will ask where a number came from, and whether the same question asked twice gives the same answer.What you end up with
- One reliable view of cash, sales, customers and delivery
- Structured, documented, queryable business data
- A data model that is yours, not a vendor's
- Answers that reconcile when someone checks them
- Stage 3 — Governed agents, not chatbotsThis is the stage we call Agent Ops. Task-specific agents do defined work against your own clean data, and every one of them has a named human approval point. An agent drafts the quote; a person sends it. An agent prepares the reconciliation; a person signs it off. An agent flags the exception; a person decides. These are not autonomous stand-ins for your staff — each has a documented scope, a record of what it did, and a person accountable for the output, which is what makes it inspectable rather than a black box nobody can explain in a data room. Where it is the right fit, this is implemented on a client-controlled, OpenClaw-based stack.What you end up with
- Agents scoped to named tasks, not general assistants
- A documented approval point for each one
- A record of what ran, when, and on whose authority
- Controls you can show rather than describe
Named deliverableThe Ownership Evidence PackHanded over with the system, not described in a meeting. It is what makes the controls above inspectable by somebody who was not in the room — an adviser, a lender, or whoever runs the operation after you.- An access map: which systems and which data each agent can reach
- The approval log format: who signs off what, and where that is recorded
- A sample of the audit trail: what ran, when, and on whose authority
- A transfer and export checklist: how to take all of it elsewhere
Why the order matters
The order is the argument: you cannot govern agents against messy data, and you cannot consolidate data through a tool nobody will stop using. Each stage is scoped and priced on its own — you are not committing to all three before you have seen the first.
Related reading
How Do You Test An AI Agent Before It Touches A Live Business Workflow?
Test an agent with an eval set of 20-50 tasks from real work, sandbox the tools, run it in shadow mode beside your current process, graduate autonomy on evidence, and turn every failure into a permanent regression test.
Your AI Agent Doesn't Have a Memory. It Has a Context Window -- And That Is the Problem
Agent memory fails quietly: context degrades answers and one poisoned write persists. What to store, what to expire, and how to test the store.
A 1.3B Model Reads Your Scans. The Interesting Part Is the 2B Model Checking Its Work
OpenBMB released a 1.3B document parser alongside a separate 2B checker that scores OCR output and flags what a human must review, and the split is what makes document automation safe near financial records.
DeepSeek Open-Sourced the Engine Behind Its AI Service. The Interesting Part Is What It Kept
DeepSeek open-sourced DeepInfer, the engine behind its own AI service, and disclosed what it earns per GPU-hour. The parts that make those economics work stayed inside the company.
Local AI or Seat Subscriptions: The Pricing Just Moved on Both Sides
OpenAI added $100 Premium seats and NVIDIA's DGX Spark rose 75 per cent in a year. What each route costs a 25-seat firm, and the register worth building before you commit either way.
Lovable Moved Its Own Site Off Next.js. The 98.5% Figure Is the Story.
Lovable has moved its own 42-million-visitor website onto the same server-rendered hosting as its 70 million AI-built apps. The migration is engineering news; the 98.5% lift in traffic from AI tools after pre-rendering is the number owners should look at.
We don't start with a diagnosis that ends in a slide deck.
A conversation about where your data actually lives, which single piece of work everything else is waiting behind, and which stage the business is genuinely ready for — before any commitment is discussed.