Rediscovery
A problem solved once is solved again because finding the first answer costs more than redoing the work.
Our purpose
Make an organisation’s best knowledge available at the moment of action.
Not stored. Not merely searchable. Present inside the task, at the point where someone is deciding what to do.
Read our thesis01
The condition
Britain does not have an ideas problem. It has an application problem.
Inside an established company, the same failure is ordinary: a hard problem solved in March is solved again in September because nobody can find the first answer.
Work is redone. Decisions are relitigated. The same lesson is bought twice. None of it appears in an account, but all of it makes the organisation slower and more dependent on the few people who remember why things work.
Variance
The same task produces a different standard depending on who happened to pick it up.
Bottlenecks
A few experienced people become the load-bearing walls of the business.
Fragility
Knowledge leaves when people do, and onboarding becomes an apprenticeship in undocumented context.
02 · The thesis
Access to frontier models is a commodity. Context is not.
Every company can buy the same capability by lunchtime. What cannot be bought is the accumulated understanding of how your organisation actually operates—its standards, precedents, exceptions and hard-won reasons.
The companies that pull ahead will be the ones that make that knowledge legible, keep it current, and wire it into the work.
Your context
StandardsPrecedentsExceptionsDecisionsThe moment of action
Relevant precedent found
Source attached
Human decision retained
Ready to act03
Working principles
How we behave on a Tuesday.
Start with the work, not the technology
Every engagement begins with a task a real person has to complete this week—with a name, a frequency and a cost.
The client’s knowledge is the product
The value is in your language, standards, exceptions and history. A generic system is the wrong system.
Humans stay accountable
Judgment, risk and client-facing commitments remain with people, visibly and by construction.
Cite or do not answer
Answers carry their sources. Where confidence is low, the system says so and routes the work to a person.
Make value visible
We establish a baseline before we build and measure against it afterwards. If we cannot say what changed, we have not finished.
Build for adoption, or do not build
A technically excellent system nobody opens is a failed implementation. Usage is a delivery metric, not somebody else’s problem.
Honest exclusions
What we will not do.
A principle that never costs anything is decoration. These choices cost us speed, scale or revenue. We make them anyway.
- 01
Replace the systems your team already runs
- 02
Hide consequential decisions inside automation
- 03
Claim a result we did not measure
- 04
Use your knowledge to train somebody else’s model
- 05
Keep ownership of what we build
- 06
Scale past our ability to keep our promises
04 · What we owe
National in ambition. Commercial in method.
There is no national productivity apart from thousands of firms doing specific work slightly better. We work at that level: one workflow, one baseline, one measured result at a time.
Capacity, not displacement
Making knowledge available to everyone gives the person in year one access to what the person in year twenty knows. That is the version of this technology we build.
Ownership, not lock-in
Knowledge bases, prompts, workflows, documentation and evaluation sets belong to the client. Renewal has to be earned by being useful.
Evidence, not assertion
We baseline before, measure after, and say plainly when the honest answer is a process change rather than a system.
The work
We are not asking Britain to invent anything new.
We are asking it to stop forgetting what it already knows.
Book an auditOne company at a time.