Skip to content
TheFrontierForge
In your tenant · Fixed price · Own it or we run it

How we work, exactly.

The methodology, the engagement model, the offer ladder, and the facts about the practice, on one page.

Book a workflow review
Check your readiness. 3 minutes, free.
The offer ladder
00The written material01The workflow review02The Production Readiness Assessment03The build

A typical engagement

Fig. 01 · A typical engagement
00Discoveryworkflow mappedWorkflow andexception map01Eval set + barbar agreedGraded eval setand the bar02Buildclears the barWorking workflowbehind review03Handoversigned offDocs and trainingor a run retainer

Scroll to view the full schematic

Each stage passes a gate and leaves you an artifact you keep, whoever runs the system next.

Durations are set per scope in the one-page scope, and a founder-attested walkthrough publishes here when it exists.

The engagement model, as fact

In your tenant
The system is built in your cloud, under your keys; nothing is copied to ours.
Human review
A human approval gate sits before anything touches a system of record.
Audit trail
Every action the system takes is logged and attributable.
Fixed price, staged payments
The price is agreed before the build starts and does not move.
Own it or we run it
You take the system, or the same team runs it; both are welcome, neither is pushed.

The offer ladder, in full

0
The written material
Everything on this site: the methodology, the engagement model, and our data posture, free to read and take to your team. The risk to you is reading time.
Free to read · reading time
1
The workflow review
A call about one workflow that hurts. You leave with a one-page scope and a fixed price, and both are yours whether or not you continue. Free; if the fit is wrong, we say so on the call.
Free · we say so if the fit is wrong
2
The Production Readiness Assessment
A paid diagnostic of the workflow against your real documents and systems: what production takes, where the exceptions live, what the bar should be. A fixed $2,500 USD, credited to the build, so it costs nothing extra if you proceed. If the assessment says the workflow is not worth building, that is what it says.
$2,500 USD · credited to the build
3
The build
The build itself: fixed price, staged payments, in your tenant. You pay by stage as the system becomes real, never the whole price up front. At the end you own it, or we run it.
Fixed price · staged payments

Fair questions

How do we start?

Read this page, then book a workflow review: one call about one workflow, and you leave with a one-page scope and a fixed price. If you would rather write first, the contact page reaches the same people.

What do you need access to?

The least that makes the work honest. Discovery runs under your access rules: read access to the workflow's documents and the systems it touches, granted by you, time-bound, and revocable. Nothing is copied out of your tenant; the full posture is on the data and security page.

How do IP and ownership work?

You own what we build for you. Code, eval suites, configuration, and documentation are assigned to you in the engagement contract, in writing. Own it or we run it refers to who operates the system, never to who owns it.

Where do your numbers come from?

Every number on a proof surface here carries a small numbered chip naming its basis. Verified means a primary source you can follow or a committed eval run; attested means the founder put their name and a date to it in a change-controlled record; pending means a declared measurement that does not have a result yet. A claim that cannot carry one of those bases is removed rather than published.

The point of view

Why most AI pilots stall, and why implementation rather than model choice is the hard part: the position the practice is built on, in the founder's words.

Fig. 02 · Position paper

The model was never the bottleneck

Every stalled AI pilot arrives at the same meeting. The demo worked. Someone senior saw it and liked it. Then months passed, and the thing is still not doing any real work. And almost always, the first idea in the room is to try a better model.

It is the wrong instinct, and it is an expensive one, because it feels like progress. Swapping models is quick, it is measurable in a benchmark, and it lets everyone avoid the harder conversation. Six weeks later the new model is in place and the workflow is still not in production, because the model was never what was stopping it.

Start with what is not in dispute. The capability is real. Today's models read a messy invoice, follow a multi-step instruction, and write a defensible summary better than the tools most firms had two years ago, and they keep improving. If raw capability were the constraint, the last three years would have produced a wave of quietly automated back offices.

An excerpt. Read the full piece

The methodology, in full

How a workflow gets from a demo into production: four stages, and the seven gates it clears on the way.

The methodology

The Seven Gates

Nothing goes live until it clears all seven gates.

The four stages

  1. 01Scope A free workflow review, then the paid assessment against your real documents and systems. The accuracy bar is agreed in writing and the eval set is built here, before any code is written.
  2. 02Build The workflow is built in your tenant, under your keys, on your own documents. The exception path and the review queue are part of the build, never a later fix.
  3. 03Prove The workflow is measured against the agreed bar on the shared eval set, with the human review gate in place. The seven gates are checked here, and a gate that does not clear stops the release.
  4. 04Hand over, or operate You own what we built: the code, the eval set, the configuration, and the documentation. Take it in house, or the same team runs it on a retainer.
Fig. 03 · The seven gatesWhat production requires
  1. 01Data and access Production needs a stable, permissioned path to the source data, with the access model and refresh cadence settled before the build starts.
  2. 02Accuracy bar and evals Production needs a written accuracy bar and a repeatable eval set that every change must pass before it ships.
  3. 03Human review Production needs a human review step with a real queue, so low-confidence cases route to a person instead of failing quietly.
  4. 04Integration and write-back Production needs write-back into the system of record, not a copied answer, with the failure path handled when the write is rejected.
  5. 05Observability and ops Production needs logging, tracing, and alerting on the live system, so a regression is caught before a user reports it.
  6. 06Governance and security Production needs the access, retention, and audit posture set in your tenant, under your keys, before real data flows through it.
  7. 07Ownership and cost Production needs a named owner and a cost ceiling per run, so the system stays affordable and maintained once the launch team moves on.

The free readiness check scores your workflow against these same seven gates. Check your readiness

The team, as fact

Built by two engineers, seven years each, with backgrounds spanning CentraleSupélec, an Oxford mathematics MSc, Amazon, and Docker. This page carries no names, no photos, and no titles by design: you meet us on the call.

Fig. 04 · Entity

References available on request.

UnchartedEdge LLC. A limited liability company registered in Wyoming.

Contact: admin@thefrontierforge.com

Practice · ShipShipOne workflow, into production, on your stack.ReadDocument · TrustData and securityHow your data is handled while we work.ReadWriting · BlogThe blogNotes on putting AI into production.Read

Reading this page was rung zero; the workflow review is rung one.

Book a workflow reviewCheck your readiness. 3 minutes, free.
  1. Book a workflow review
  2. Commission a Production Readiness Assessment
  3. Commission the build