AI as an assistant
AI as a production line
Autocomplete speeds up one engineer. A factory changes what the team can take on — and keeps working when that engineer is on holiday.
Dark Factory
Automate the entire SDLC. Without lowering the bar.
A dark factory runs lights-out — the floor is automated and people set the standards. I help enterprise teams build that for software: from requirements and architecture through agent rules, automated review, testing and production deployment.
- Years in the industry
- 15+Years in the industry
- Automated SDLC stages
- 7Automated SDLC stages
- Security & compliance
- EnterpriseSecurity & compliance
01The thesis
Generating code was never the hard part.
Enterprises can already produce code faster than they can review, secure, test and operate it. The bottleneck moved. A Dark Factory addresses the whole line — not the keystrokes.
Standards in a wiki
Standards in the repository
Conventions nobody reads get ignored by humans and agents alike. Rules that live next to the code are applied on every single change.
Review as a bottleneck
Review as a gate that scales
When agents produce more diffs, human review becomes the constraint. Automated reviewers triage first so people spend attention where it counts.
02The pipeline
Seven stages, each with an artefact you can inspect.
Every stage produces something durable — a document, a rule, a suite, a pipeline. That is what makes the factory auditable instead of magical.
- 01
Requirements
Product requirements gathered and written so an agent can act on them. Ambiguity is the most expensive defect in an automated pipeline — it compounds at every stage downstream.
- PRDs
- User stories
- Acceptance criteria
- 02
Architecture
System, data and integration architecture decided deliberately and up front, then recorded — so every later decision has a reference point instead of a guess.
- ADRs
- C4 diagrams
- Data model
- 03
Agent rules
Markdown rule files that encode your standards — structure, naming, boundaries, error handling, security. Agents generate code that already fits the codebase, rather than code you have to rewrite.
- CLAUDE.md
- Rule files
- Skills
- 04
AI code review
Specialised reviewer agents check correctness, architectural boundaries, security and regulatory exposure — before a human ever opens the diff. Humans review the findings, not the whole surface.
- Reviewer agents
- Security checklist
- Review records
- 05
Test generation
AI-written code gets AI-written tests, held to genuine coverage and quality standards. A green checkmark is not the goal; a suite that fails when the behaviour breaks is.
- Unit suites
- Integration suites
- Coverage gates
- 06
E2E & load
End-to-end journeys and load profiles that prove the system behaves under production conditions and production traffic — not just on a developer's machine.
- E2E journeys
- Load profiles
- Performance budgets
- 07
Production
Pipelines, environments, observability and rollback. The part most AI demos skip, and the only part your customers actually experience.
- CI/CD
- Observability
- Rollback
03Standards
Speed is only useful if the result holds up in production.
Enterprise software carries obligations that a prototype does not — security, regulation, uptime, and the ability to explain how something was built. The factory is designed around them.
Built for enterprise security
Injection risk, secrets handling, data exposure and PII classification are review dimensions in the pipeline, not an audit you fail later.
Standards that survive scale
Rules live in the repository, versioned and reviewed like code. They apply identically to every engineer and every agent on the team.
Auditable by design
Every stage leaves an artefact — requirement, decision record, rule, review, test report. You can show how a line of code came to exist.
Your team owns it
The factory stays after I leave. Teams are trained to extend the rules, not to depend on me to maintain them.
04Track record
Fifteen years of shipping software the enterprise depends on.
Long before the current wave — .NET, React, cloud architecture, e-commerce platforms and consulting. The standards came first; the automation is what changed.
- Ritech
- Hermex
- FullStack 1001
- SEI Consulting
- Teerana
- Dicom (IVEH)
Kreshnik is good - really good. Best developer that I've ever worked with. He's creative and works hard on the details.
Working with Kreshnik was an absolute treat. He explained parts of the process as he went on, made sure to include me in all of the decisions, and always maintained an amazing level of communication with me.
Kreshnik understood our request and needs well. He presented multiple solutions beyond requested tasks.
05Contact
Tell me what you're building.
Whether you're starting a product or trying to get an existing team to move faster without breaking what already works.
- A short call to understand the product and the constraints you actually have.
- An honest read on which stages of your SDLC are worth automating first.
- A plan you can act on, whether or not you work with me.