Why Trilithos

The hand-off was work.Nobody counted it.

A firm that advises clients and then builds for them does three distinct pieces of work, not two. Trilithos exists because the join between them was where the work was being lost.

01Three activities

Consulting. Building. And the hand-off between them.

  1. 01

    Consulting

    Understanding a business well enough to recommend a direction. The report is written for the client's board: what should change and why — not how a system should work.

  2. 02

    Building

    Software cannot be built from statements of intent. Every field, rule and exception must be defined first.

  3. 03

    The hand-off

    The report must become a specification, and the developer doing that did not attend the client meetings. Every point the report left open is decided without the client seeing it. It is the activity nobody counts.

02The gap

What happens between understanding and building.

A consulting report may recommend automating purchase approvals. It will not state the approval limit for each role, what happens when an approver is on leave, or how an exception is recorded.

  1. 01Client conversation
  2. 02 · AItlasBusiness understanding
  3. 03The gap03 · Data WeaverApproved requirement → build contract
  4. 04 · AnvilSoftware build

The report is reinterpreted by someone who was not in the meetings.

  • Duplicated understandingConsultants learn the business and write the report; developers then work the business out again from that report.
  • DelaysConsulting and building run as two sequential projects joined by a person, and work queues behind a single long task.
  • Developer questionsBuilds stall while developers wait for answers only consultants can give.
  • ReworkCorrecting a system that did not match what the client intended means repeating part of a delivery.
  • MismatchWhat was built differed from what had been agreed.

03Why AI

None of it could be solved by working faster.

What a client agrees to is written in ordinary English. Until now, only a person could turn that into something a developer could build.

That conversion — from agreed business language into a precise, buildable specification — is the operation at the centre of Trilithos. It is the reason the platform is built on language models rather than on forms and templates.

04The principle

AI proposes. People decide.

This is not a courtesy or a disclaimer. It is enforced structurally: the platform stops and waits at defined points, so approvals are recorded rather than assumed.

  • ConsultantAItlas produces the analysis. The consultant reviews and corrects it, and their edits take precedence over the model.
  • DeveloperAnvil recommends the technology for each feature and states its reasons. The developer accepts, changes or replaces any of it.
  • DeveloperThe feature list is editable, and the edited list — not the generated one — is what gets built.
  • ClientThe client approves a working preview of the whole application before any of it is built.
  • EveryoneSignificant decisions stay with people. Every phase of a build, and every deployment, is started by a person.
Where the model decides, and where it does not
The model decidesA person decides
What the research saysWhether the research is right
Which capabilities to recommendWhich capabilities to keep
What technology to propose, and whyWhich technology is used
What features to draftThe feature list that gets built
How a screen is composedWhether the design is approved
How code is writtenWhether a phase proceeds
Whether a check passedWhether the application is released

05Grounded output

AI that knows when it doesn't know.

Constraining a model turned out to be harder than generating output from it. These controls are product features, and they are why the output can be put in front of a client.

  • NeverInvent a fact to fill a gapInsteadLeave the field empty. A client who finds one invented fact will doubt the whole document.
  • NeverPretend to knowInsteadState what is not known, and stop.
  • NeverBlend inference with evidenceInsteadKeep hypotheses separate, each with its reason, adopted only by explicit action.
  • NeverLose analysis on the way to the clientInsteadEvery generated field reaches every output format — a test fails the build if one does not.
  • NeverCall an application delivered before it runsInsteadAsk whether it is running, and release only once that is confirmed.
  • NeverShow a green step when nothing was verifiedInsteadSay exactly that: nothing was verified.

06What makes it different

Built as one structure, not assembled from tools.

  • 01

    One continuous system

    Business understanding flows into software delivery. Consulting and building are no longer two projects joined by a person.

  • 02

    The hand-off is explicit

    Data Weaver is a named platform with stages, tests and failure states, preserving the understanding the client approved.

  • 03

    Human decisions stay authoritative

    AI recommends; professionals decide. Technology, features, design approval, each build phase and release belong to a person.

  • 04

    Grounded output

    The system refuses to fill gaps with plausible content, and says when it could not ground or verify something.

  • 05

    Standard software

    Applications are built on mainstream technology the developer selects, which the client's own engineers can maintain.

  • 06

    Not locked to one AI provider

    Nine model providers sit behind one configuration setting, and either platform can run inside a client's own infrastructure.

07Who Trilithos is for

Everyone on the path from advice to software.

  • 01

    Consulting teams

    Spend more time making decisions and less time assembling delivery artefacts. Every engagement includes the same five analyses.

  • 02

    Consultants

    Review, correct and guide the analysis the platform produces, and keep authority over the engagement — without writing code.

  • 03

    Developers

    Start from an agreed build contract instead of reverse-engineering a consulting report — and keep the decisions on technology, features and phases.

  • 04

    Business leaders

    See one continuous process from the business problem to a released application, with approvals recorded at defined points.

  • 05

    Clients

    See what is proposed, approve a working preview before the build starts, and receive software only once it is confirmed running.

08Plainly stated

What Trilithos is not.

  • Not a replacement for judgement

    Client interviews, judgement and relationship work take as long as they always did.

  • Not a no-code tool

    It produces standard software on mainstream technology that ordinary engineers maintain.

  • Not a lock-in

    The developer selects the technology from mainstream options; the client's own engineers can maintain the result.

  • Not tied to one AI provider

    Nine providers are available through a single configuration setting.

Enter the platform

Two doors, one structure.

Consultants work in AItlas. Developers work in Anvil. Data Weaver carries the approved requirement from one to the other.

Data WeaverHand-off inside Trilithos · approved requirement → build contract

For consultants

AItlas

  • Understand the business.
  • Build the strategy.
  • Define what should be built.
Open AItlas

For developers

Anvil

  • Turn the approved requirement into software.
  • Design. Build. Deploy. Release.
Open Anvil