The AI Engine

The AI Engine is how we build. Now it can build for you.

The Engine is how we deliver software now: five AI Kits of working agents, the process that runs them, and 25+ years of hard-won judgment about what breaks in production. It's our evolving in-house framework built on real projects. When our people plug in to your project, the whole Engine comes with them.

Through every project, the Engine grows smarter so the next one starts further along.

Talk to our team

Inside the Engine

01 Specification Kit → 02 Archaeology Kit → 03 Agentic Coding Kit → 04 Data Team Kit → 05 Cyber Kit →

10 days

from a legacy codebase to a working modern application with dozens of screens, and none of the old code

4.8B

data points a trend viewer we built in weeks can chew through

50%+

less delivery effort with AI-first methods, by our own estimate

years building and running systems people cannot afford to lose

Five AI Kits. You take the ones your business needs.

Each kit is a working set of agents, templates and checks for one part of the development. Together they make up the Engine. We bring what your case calls for and leave the rest at home.

Specification Kit

Turns what you want into a specification precise enough for an agent to build from, instead of leaving it to fill in the blanks itself.

Archaeology Kit

Reads an old system and works out what it actually does: code, database, configs, screens, tickets. What used to take months of interviews and digging, now facilitated by AI.

Agentic Coding Kit

Agents that write, review and test code side by side, one epic per run, with your architecture and coding rules built into the run itself.

Data Team Kit

Lets one data specialist carry a request from question to finished product, instead of five people handing it down a line and losing the context.

Cyber Kit

Threat detection, security analysis and compliance checks, running in the same pipeline as the build rather than after it.

Two things make it work.

Specs, not chat

A chat thread is a poor contract. Nobody can tell later what was agreed, and an agent will happily guess at the gaps. So we write the requirements down in a form both people and agents read the same way, and keep the architecture rules and coding standards in source control next to the code.

Every feature points back to a line in that specification. The audit trail is a side effect, not a separate project.

Swarms, not one big prompt

One agent with a long prompt gets you a demo. What we run instead is a set of small agents with defined roles, working in parallel, each with its own slice of context, and separate agents whose only job is to check the ones doing the writing.

People set the direction, hold the gates and sign things off. That part we do not automate.

One kit at work

What the Archaeology Kit does to a system nobody understands.

An old application, no documentation, the people who wrote it long gone. Here is the run, stage by stage. It advances on its own, or you can click through it.

Stage 01

Point it at everything

Stage 02

Agents read in parallel

Stage 03

You get facts, not guesses

Stage 04

Straight into the build

Everything counts as evidence

We hand the Archaeology Kit whatever exists: the source, the database, the config files, the logs, the screens, even the ticket history. Nobody has to write a summary first, which is the step that usually stalls for a month.

Access and data boundaries are agreed before anything reads a line.

230k lines of source Stored procedures Config and deploy files Runtime logs UI screens Ticket history

Three passes at once

Separate agents take separate questions and run at the same time. One follows the dependencies, one digs out the business rules, one traces what a user actually does end to end. A fourth checks their work against the code.

An architect reads the findings as they land and redirects the run when something looks wrong.

Dependency and integration mapagent 1

Business rules and domain logicagent 2

End-to-end user flowsagent 3

Verification against sourceagent 4

A factual account of how it behaves

Not an opinion about the system, a description of it, derived from the thing itself. Including the flows the people who own it no longer fully see.

The open questions get written down too. Those are the ones we bring to your business owners.

Architecture map

What talks to what, and through which door

Data model

The real one, as built, not as drawn in 2011

Feature map

Every behavior found, traced to where it lives

Open questions

Rules only a human can settle

The findings become the spec

This is where the Archaeology Kit hands over to the Specification Kit, and then to the coding agents. What the old system does becomes the contract for what the new one has to do, with your business owners deciding what changes on the way.

Nothing is thrown away between stages, which is why the whole thing takes days rather than quarters.

01Legacy-anchored specification, reviewed by your owners

02Boundaries set: what changes, what stays exactly as it is

03Coding agents build against that spec, in parallel

04Engineers review, then it ships with a name against it

People at the edges, agents through the middle.

Seven phases from idea to production. The bar shows how much of each phase the agents carry. Nobody skips the gate at the end of one.

Ideation

We agree where this is going before anything gets built.

10% AI · gate: direction

Planning

That direction turns into a backlog somebody has signed.

30% AI · gate: spec sign-off

Design

Specs you can run and prototypes you can click, not slideware.

50% AI · gate: architect

Development

The swarm builds in parallel, one epic at a time.

90% AI · gate: merge approval

QA

Quality is proven before a human is asked to look.

70% AI · gate: UAT

Launch

Cut-over, then hypercare. Quiet is the goal.

40% AI · gate: go-live

Maintenance

What production tells us goes back into the backlog.

60% AI · gate: release

Fast is only useful if it holds.

The Engine ships with the boring parts: the logs, the reviews and the sign-offs. Ask and we will show you the evidence.

Every call is logged

Model calls are tied to your engagement, checked against spend and model limits, and written down: which model, how many tokens, what it cost, what came out.

Reviewed, not just generated

Architects set the boundaries, engineers check the output, testers hold the line. Nothing reaches production without a named person putting their name on it.

Your code stays yours

We work on enterprise and business plans, with data encrypted in transit and at rest. Nothing you give us is used to train a model, and retention is capped by policy.

Related AI-first capabilities

The Engine sits underneath everything below. Start wherever your problem is.

AI-driven legacy modernization

What the Archaeology Kit does to a system nobody has documented.

AI-first software delivery

How the method changes the speed and the economics of a software project.

AI-native teams

The teams that run the Engine, brought in without you hiring for it.

Jarkko Enden

CTO, Finland

Tanel Tensing

Group Head of Engineering

Evgenij Nazarenus

Head of Data & AI, Germany

Kurt Knaub

Head of Data & AI, Americas

Allan Bernard

Head of Engineering, Estonia

GET IN TOUCH

Let's discuss your AI Engine needs.

Tell us the outcome you need. Meet with our experts to work out which kits apply, what they would change in your delivery, and what it would take to get started. Bring a system that worries you, a deadline that looks impossible, or just a question about how any of this works in practice.

Plain answers to the jargon.

The vocabulary around AI delivery is a mess. Here is what we mean when we use it.

What is the AI Engine?

Everything we bring to a software project now, working as one thing: the five AI Kits, the delivery process that runs them, the gates and reviews that keep it safe, and 25 years of accumulated judgment about how systems fail in production. The kits are the visible part. The Engine is the kits, plus the method, plus the people who know when to override both.

What is an AI Kit?

One of the five working sets of agents inside the Engine: Specification, Archaeology, Agentic Coding, Data Team and Cyber. Each holds the agents, prompt patterns, templates and guardrails for its own part of delivery. Built on real projects by the people delivering them, and a little better after each one.

What is an AI POD?

A delivery team made of our engineers and AI agents, scoped and priced before we start. The Engine is what the pod runs on. Pods plug in for the job and unplug when it is finished.

What is AI-native engineering?

Building with agents as the default way work gets done, rather than as an assistant bolted onto an old process. The difference between AI-assisted and AI-native is the difference between giving people better tools and changing how the work flows.

What is spec-driven development?

Writing requirements as a contract a machine can read, instead of instructions in a chat window. The spec is the single source of truth and the definition of done, so agents build what is defined and every feature traces back to a line in it.

What is agentic swarm coding?

One agent coordinating a group of smaller, specialized ones that work at the same time, with separate agents checking what the others produce. Tasks move in parallel instead of queueing behind one developer, under human orchestration and gates.

What is legacy archaeology?

Pointing agents at a system nobody fully understands any more, including the source, the database, the configs, the logs and the screens, and getting back a factual account of how it behaves, plus specs anchored to that.

Will it work with our stack?

Yes. The Engine is tool-independent. We work with what you already run and pick the agents, ours or someone else's, that fit your landscape, your compliance regime and your data boundaries.

Do we get to keep it?

A version of the Engine tuned to your estate stays with you, kits included, along with whatever archaeology recovered about your old systems, written down. Your team works the same way we do, so the capability does not leave when we do.