How we work

Seven stages from idea to a system that scales

The sequence does not change with the size of the engagement. What changes is how long each stage takes, and who on your side needs to be in the room.

The stages

Idea to production, in order

Architecture is decided before code. AI and agent design is decided before the application is built around it. That order is what survives production.

StageWhat happensWhat you leave with
01 · IdeaThe business process, the users, the data that exists and the constraints that are not negotiable.A written problem statement we both agree on
02 · ArchitectureTenancy, data model, integration surface, security posture and where the system meets the ones you already run.An architecture decision record for every non-obvious choice
03 · AI / agent designWhere agents, retrieval, tools and deterministic workflow each add value, and where they do not. Tool grants and human gates decided here.A held-out evaluation set and a definition of “accurate enough”
04 · DevelopmentThe application, APIs, agent workflows and data layer, built as one system with tests from the first week.Working software at the end of every sprint
05 · IntegrationsCRM, ERP, payment, identity and the systems of record, through explicit, permissioned, logged contracts.Every external effect replayable and revocable
06 · DeploymentInto a production environment on infrastructure held in your own accounts, with a rehearsed rollback.A system you own outright
07 · ScalingPerformance, cost, reliability and business outcome, monitored after go-live, with the AI layer instrumented like everything else.A running view of what it costs and what it earns
Who you are working with

A product engineering company, not an agency

Founded in 2019. One team covering design, engineering, quality analysis, DevOps and cloud, under one accountable lead. What changes between a founder and an enterprise is scope and pace, not the standard of the engineering or who does it.

Startups · idea to MVP

A first version that proves the approach on real data before a runway is spent, then the engineering that keeps it standing when usage arrives. No-code where that is the honest answer.

Growing businesses · automation and integration

Process automation and decision support built into the systems your team already uses, with approval steps where they belong and nothing that fails silently.

Enterprise · complex systems

Multi-tenant platforms, agent layers on live systems, and modernisation that does not require a rewrite. Access control, audit trails and human gates designed in.

For startups

Get to something real before you spend a year building it

MVPs, automation and first AI features for companies that need proof before they need scale.

No-code MVP

n8n and Replit. Working automation or an internal tool in days rather than sprints. Four of these are public repositories you can read.

Coded MVP

A narrow slice built properly, with a held-out evaluation set that proves the model earns its place before you invest further.

Your first AI feature

One capability into an existing product, at one seam, reversible. If it does not earn its place it comes out cleanly.

Fractional CTO

Technical judgement, architecture ownership and hiring help, without the permanent hire.

The honest version: most first-year products should start no-code and move the parts that earn it into code. Starting with a full build is how a runway gets spent before anyone has learned anything.
Consulting & advisory

Advice before, or instead of, a build

Four ways to get our judgement without commissioning an engineering team. Every engagement is scoped to the problem in front of it, so there is no price list.

EngagementFormatWhat you leave with
Architecture review60 minutes, no chargeOur read on where the risk sits and what we would decide differently, in writing
AI readiness assessmentA short engagementWhere AI genuinely applies, what it costs to run, and what to build first: architecture, roadmap and cost model, all yours
Technical due diligenceScoped to the codebaseAn independent read on a codebase, a team and a technical claim, for investors, acquirers or an inheriting team
Fractional CTOOngoingTechnical leadership, architecture ownership and hiring judgement, without the permanent hire

Tell us what you are trying to decide. A paragraph on the system or the decision in front of you is enough; you will get a reply from a person who builds, not a sales sequence.

No-code, or code?

We will tell you when no-code is the right answer, and when it is not

Much of what a company needs in its first year does not need a codebase. Some of it does, and building the wrong one is expensive. We have published both kinds of work.

QuestionNo-code is right when…Code is right when…
VolumeTens or hundreds of runs a dayVolume makes per-run cost or latency matter
LogicThe workflow is linear with clear branchesState, concurrency or transactions are involved
DataIt can live in the toolYou need your own schema, migrations and history
OwnershipYou accept a platform dependencyThe system must outlive any vendor
MoneyNothing settles or reconcilesAnything touches payment, ledger or identity
AI LinkedIn Content Manager. Seven chained agents, each with its own structured output parser, behind an approval gate that pauses the run and resumes on a person’s decision.
Evidence

Four public automations

Built on n8n, each with its own landing page and repository. These are the no-code half of the practice, and they are open.

LinkedIn Content Manager

Seven chained agents, seven output parsers, one human approval gate. 44 nodes, public.

Restaurant Automation

Files supplier paperwork from email and reports daily POS sales. Five distinct logged failure branches, plus a nine-node mock POS so the client could watch it run before handing over credentials.

Business Doc Agent

One form submission becomes finished documents. Validation before any model call. MIT licensed.

Blood Report Analysis

A PDF or photograph becomes a plain-language summary by email. Multi-modal intake with a separate errors sheet, so nothing fails quietly.

Security & ownership

What we build to, and what stays in your name

Repositories, cloud accounts and vendor accounts are yours from day one. There is no lock-in, and nothing runs anywhere we control.

ISO 27001 practice

One platform we engineered was taken through ISO 27001 certification with the client. PremitiveKey is not itself certified, and we state that plainly.

GDPR and SSO

Data handling designed for GDPR. Employees sign in with existing company credentials through client SSO, so there is no password rollout and no list to leak.

Client-owned everything

Repositories, cloud accounts and vendor accounts in your name from day one. Switch us off and nothing stops.

Engineering controls

Access control, secrets management, encryption, backups and append-only audit trails, as standard rather than as an afterthought.

How an engagement starts

Three steps, and the first two cost nothing

The way to find out whether we are the right team is to put a real problem in front of us and read what comes back.

STEP 01 · FREEA 60-minute architecture review

Bring the system you have or the one you are planning.

STEP 02 · YOURS TO KEEPThe notes, in writing

What we would build, what we would not, and why.

STEP 03 · ONLY IF IT MAKES SENSEA scoped engagement

A first agent, an MVP, or a platform. If the review points elsewhere, we say so.

Start with one conversation

Sixty minutes on the idea, the system or the decision in front of you. No charge, and the notes are yours either way.

A 60-minute architecture review · no charge · the notes are yours either way