HomeHow We DeliverOur WorkWorking With UsCompany

How we deliver·Governed agentic delivery

AI speed you can actually trust.

Working software. A complete record. Quality that compounds.

SPEED → GOVERNANCE / TRUST → Traditional vendors discipline, without speed AI-first shops speed, without discipline Meta Functions fast, trusted, and climbing every cycle
How work moves

One delivery line. Two disciplines. Every gate named.

Four lines run through every stage — the AI, our team, your team, and the guardrail that never switches off. All of them meet at a named gate before the work moves on.

Stage 00 of 08

Engagement & Scoping

See it work
The AI Our team Your team Guardrail The gate
Always visible to you sprint demos weekly status progress dashboard delivery ledger Ownership shown is indicative — the exact split depends on your engagement model.
See it workTrust it furtherMake it yours ↻ one maturity rung higher each cycle

1Why we work this wayReasoning

Why we work this way.

For twenty‑two years, quality depended on individuals who vary, rotate, and forget. Then AI moved the bottleneck: writing software is no longer the hard part — trusting it is. So we built a system that makes quality automatic and trust checkable, from day one.

Small, reviewable steps

Every change is small enough to review on its own — nothing moves in one large, unreviewable leap.

A named person approves

Nothing merges without a named person approving it — no change ships on an AI's say-so alone.

Every AI touch, on record

We know exactly what the AI touched — and what it didn't. Guardrails fire at the boundary and cannot be switched off.

Audit-ready by default

Evidence is generated as work happens, never reconstructed. Findings become standing rules — 35% to 64%, and rising.

How we got here

The short version of our story — from decades of engineering discipline to the practice every engagement runs on today.

Twenty‑two+ years in

Discipline, learned the hard way

Banks, engines, ships, railways — quality by review gates, named accountability, evidence.

Then AI arrived

A new kind of speed becomes possible

AI writes software faster than teams can verify it. Guardrails that rely on the AI's cooperation prove to be none at all.

So we shaped a practice

A disciplined way of working, not another assistant

Sealed AI roles, human gates, and a record generated from events — industrial-grade quality discipline, applied to software delivery.

Today

Every engagement runs this way

First-time acceptance has climbed from 35% to 64%, cycle over cycle.

Part of every engagement — ready from day one

Ready before you start.

Four accelerators, built from twenty‑two years of delivery experience — ready on day one, applied to your engagement alone.

The Role Library

A deep library of pre-scoped AI capabilities — ready on day one, tailored to your context.

The Standards Ledger

Engineering standards distilled from decades of practice — applied as standing rules from the first commit.

Boundary Policies

Guardrail policies grounded in deep regulated-industry experience — banking, aviation, maritime, infrastructure — active from day one.

Cycle Templates

Planning and estimation patterns drawn from years of disciplined delivery practice — dependable from the first cycle.

2How the delivery cycle worksPrinciples

Not a copilot — a governed delivery team.Not a copilot. A governed team.

In plain terms: a factory line for software — AI moves the work, a named engineer opens every gate, and the line stops anything outside the rules. Here's what that looks like, cycle by cycle:
01
We ask first

Learns your world. Asks — never assumes.

doc ingestioncontext mapassumption log
02
You approve the plan

Itemised, approved line by line, before anything runs.

itemised planestimateshuman gate
03
We build in small steps

Every change reviewed and tested; a senior engineer holds the merge.

sealed agentsauto review + testsboundary guard
04
You get the proof

Proof from the record; findings become next cycle's rules.

proof generatorimmutable recordrules distilled
step — click or hover for the internalsgate — a named person approves
under the hood · ask
ingest47 documents read · standards, policies, domain notes
mapentities and constraints wired into a context map
ask12 clarifying questions raised · 0 assumptions made
plan14 items · effort estimated per item
gatehuman approval required — line by line, your side
lockthe approved plan becomes the contract for the cycle
agentscontract-scoped roles · least-privilege access only
verifyauto review + tests on every single change
guardboundary controls fire whatever the agent decides
gatea senior engineer holds the merge
proofpage generated from the immutable record
honestalways includes what is not yet done
learnfindings distilled into standing rules for the next cycle

Three more guarantees the cycle above doesn't show on its own:

Guardrails built in

We say what it can and can't touch.

They cannot be switched off — by anyone, including us.

Transparent cost

Your own AI subscriptions.

Any cloud, or an in-country model. Metered, capped, no markup.

A full team, not one bot

Specialist AI depth, every discipline.

Each sealed behind a contract defining exactly what it can see and do.

3The climb to expert practicePractice

Most teams get stuck on ad-hoc wins. We build past it.Most teams stall. We build past it.

From scattered AI experiments to a practice that compounds, cycle over cycle. Click a level to see where teams get stuck — and where we take you.

Level 02 — decaying, needs escape velocity
Level 05 — locked, sustained without us
4How the three approaches compareComparison

Five axes, one shape that holds.

Most delivery approaches are strong on one or two axes. Ours doesn't give any of them up.

5The difference, side by sideCompared

The same project, two very different experiences.One build. Two outcomes.

Without governed delivery With Meta Functions
01Week one Kick-off decks, no builders Asks, never assumes
02The plan Verified much later Approved line by line
03While building Reported, not visible Visible as it happens
04Goes wrong Surfaces late Blocked at the boundary
05Delivery Summary, on trust Proof from the record
06Next project Lessons leave with team Starts from what's learned
6Every action, on the recordAudit Trail

Audit-ready by default.

Every change writes its own ledger entry the moment it happens — not reconstructed after the fact.

Delivery ledger
0 entries logged
Change
Timestamp
Size
Approver
Gates
Status
FOR TECHNICAL READERS
7One engagement, traced stage by stagePipeline

What's underneath.

Everything downstream ships into your own systems.

03 / 05

8The twelve-factor model, extendedFactors

Fifteen foundations, not a blank page.

Runs reliably anywhere, connects cleanly with the systems you already have, and stays accountable from day one. In engineering terms: cloud-native, API-first, governed by design.

01 / 15
9The gates in the middle never changeGates

Every path, the same door.

A one-line fix and a new feature start differently. They go through identically — so nothing small enough to feel safe skips review, and nothing large enough to be risky finds a shortcut around it.

Ask
Approve
Build
Proof
Delivered
> choose a request type above
10How the layers separateLayers

Built for encapsulation. Enforced at the boundary.Built for encapsulation.

system layers
└── PEOPLE — the console where humans decide
    └── AGENTS — sealed specialist roles: a contract, rules, knowledge, tools, typed skills
the boundary — controls fire here, whatever the agent decides
└── CONTROL — every call brokered · record immutable · spend capped by you
    └── SYSTEMS — a name, a scope, a pointer. never the credential itself.
Inside the boundary
Draft code changes
Bounded to the files and scope named in the request.
Run tests in a sandbox
Never against production data or live systems.
Propose config changes
Proposed only — a named person still approves them.
Outside the boundary
Production data
Never read, written, or used to train anything.
Credentials & secrets
Not visible to the AI at any stage of the cycle.
Direct production deploy
Every release goes through the same named-approval gate.

Guardrails that rely on an AI's cooperation are not guardrails. Ours fire whatever the agent decides — and cannot be switched off, by anyone.

For the CIO

Speed you can defend

Working software in weeks, with evidence at every step — and parallel pods when the scope calls for it.

For the CFO

Two clear costs

A delivery fee agreed for the engagement scope, and AI usage on your own subscriptions — no markup.

For the auditor

Evidence already exists

An immutable record and a proof page per cycle. The answer is an export, not a scramble.

For the engineering team

Capability that stays

Code, docs, and the record are handed over from the first commit; the in-house team is certified to direct the work.

11Your window into deliveryConsole

Your product is what you receive. The console is how you watch it being built.The console shows you everything.

Your team oversees the work through the delivery console — live, throughout.

plan approval · cycle view
07 · Add tender-scoring export
build role · est. 3 units
Approve
08 · Refactor officer-workflow queue
build role · est. 5 units
Approve
09 · Extend audit-log retention
data role · est. 2 units
Return
Approved 13 of 14 · client engineer · 09:52✓ gate open

The plan gate — your side approves the work, line by line

proof page · cycle close
Shipped — accepted first time7 / 11 · 64%
Spend — your AI subscriptionwithin cap · 0% markup
Quality trend35 → 48 → 64% ↑
Needed a person — the honest section2 items · with reasons

The proof page — generated from the record, never written by hand

delivery console · one governed cycle — live view● running

A governed cycle end to end · reference environment shown

Why it matters: a report with only good news is not proof. Every figure links to recorded events — including what is not yet done.

A delivery practice that arrives already disciplined — your standards, enforced from day one.

One conversation with the engineers who would build the work — email us and a senior engineer replies.