The LockedIn Method
How we work
The LockedIn Method: an agentic SDLC framework for regulated enterprises
AI agents make software output cheap. They do not make it correct, finished, installed or wanted. The LockedIn Method is how we deliver with agents so that speed does not cost control. It rests on four commitments.
LOCK
- Lock the scope. Work starts from a fixed, numbered list of requirements, each with the checks that prove it is done. Scope changes by a recorded decision, never by drift.
- Own the outcome. One accountable owner per program decides, records the reason, and answers for the result.
- Check independently. Every change is read by someone who did not write it. The person who proposes a change never approves it. Quality is rated by an evaluator who knows nothing of how it was built.
- Keep the evidence. Done means proven: a requirement counts when its checks pass and the proof is on record. Progress is one number, counted the same way every time.
Agents have identities, and limits
An agent is not an anonymous process. In our method every agent has its own identity, a named person who sponsors it, and an owner who answers for it.
- Registered and approved. An agent is registered before it works, and a second person approves it. People sign in through the organization's own identity provider; agents and services receive their own credentials, never a person's.
- Delegated, scoped, expiring authority. An agent acts with authority a person delegated to it, limited to named tools, data and budgets, for a stated time. A revoked credential is refused on its next call.
- A person decides what a person must decide. An agent can ask for an approval, a budget or an exception. It cannot grant one, approve its own request, or lift a stop placed on it.
- Stops that hold. Any agent can be paused at once. A stop on a production agent takes a second person to lift, and the person a stop covers cannot lift it themselves.
- Everything on record. What an agent did, under whose authority, with which model and tools, is written to a signed record an auditor can verify.
This applies to an agent that helps an engineer for an afternoon and to one that runs a recurring task unattended.
Three points of control, one set of rules
| Where | What is enforced there |
|---|---|
| On the developer's machine | What an agent may run, read or send, and with which model. |
| Across the organization | Who and which agent may do what; the required check before a change merges; the audit record. Policy is written, approved and published here, in one place. |
| Across the program | The fixed scope, who holds each item, the evidence that each is done, capacity, and what is built, installed and proven. |
The rules are policy as code: versioned, approved by two people, and enforced by software rather than by a document.
Inside your environment
We work within your systems and your rules. Your repositories, your pipeline, your approvals and your release calendar stay as they are. What we add sits on top: a fixed list of what is being built, proof attached to every change an agent writes, and a named person who reviewed it. The work stays inside your boundary, under your identities and your model contract, and nothing ships that your process would not ship.
Built for audit
Each rule of the method produces evidence, and each is mapped to the controls an auditor asks about. Security testing is part of the flow: code is analysed on every change, a running build is tested before it is promoted, and independent testing follows.