Lakewright AI · AI consulting
AI you can audit.
We help organisations put AI to work on their data and their code, and make sure every AI decision and every AI-made change is recorded, attributable to the request that caused it, and reversible. If someone asks what the AI did and why, you answer from the record, not from memory.
For teams that will be asked to explain their AI.
Regulated data teams
Healthcare, insurance, finance and others whose data arrives as claims files, HL7 feeds, mainframe extracts and scanned documents, and who cannot send it to someone else's cloud to make sense of it.
Engineering teams adopting coding agents
Teams where agents now write much of the code, often several at once, and where nobody can say which request produced a given line or undo one request without undoing the rest.
The people accountable for AI
Security, risk and compliance leads, and the executives who sign off, who need an audit trail behind their AI before it goes anywhere near production.
Auditable by construction, not by after-the-fact paperwork.
Everything we build and recommend follows from one rule: an AI system is only as trustworthy as the record of what it did. We design that record in from the start.
Recorded
Every AI decision and every AI-made change is captured as it happens, not reconstructed later from logs nobody kept. Lakewright records every ingestion decision; mcpc snapshots every edit an agent makes.
Attributable
Each change points to the request that caused it and the person who approved it. In Lakewright an AI proposes the plan and a person approves it; in mcpc each edit is linked to the prompt behind it.
Reversible
The unit of undo is the thing you actually regret. mcpc can revert one whole request while keeping another agent's work; Lakewright loads nothing without an approved plan you can inspect.
Auditable observability behind every AI decision, for the enterprise.
Four ways we work with you.
Two products, built on the same rule.
We build the tools we use on engagements. Both are pre-release, and we say so.
Lakewright Pre-release
Self-hosted, governed data ingestion into Apache Iceberg. Reads X12 EDI, HL7 v2, FHIR, mainframe fixed-width, scanned documents and database change streams, and lands them as governed tables inside your own environment. An AI proposes each plan; a person approves it; every decision is recorded.
About Lakewright →mcpc Pre-release · open source
Coordination and version control for AI agents: a server that agents from any vendor talk to over MCP in place of git. Every edit is linked to the prompt that caused it, a whole request can be undone, agents hear about each other's changes before they collide, and repository rules are checked before a bad edit lands.
About mcpc →You work with the people who build the tools.
Lakewright AI is a small, senior firm. There is no sales team and no hand-off to juniors: the people who scope the work do the work, and build the two products it runs on.
For the thinking behind mcpc, read the essay We replaced git with MCP.
Most engagements start with one conversation about one problem.
Tell us what you are trying to do: a data estate nobody has mapped, agents you want to trust with more of the code, or an AI rollout that needs an audit trail. We will say plainly whether we can help.