Your product, end to end
Architecture, implementation, tests and deployment for a web product or an API. The build a software house quotes in months, run by one engineer and a crew of agents.
ArchitectureFull stackCI and testsDeployment
One engineer, a crew of agents · Kraków, working remotely
You don't need a software house.
You need one engineer running a crew of agents.
I'm Krystian. Ten years of shipping product software, and now a crew of AI agents doing the volume while I stay accountable for what lands: the architecture, the review, the tests and the way it behaves in production. Vibe coding gets you a prototype. This gets you something you can run.
Delivery run by agentsSenior review on every changeNo team to keep fed
00briefscope and acceptance agreed
01agentsimplementation, migrations, tests
02reviewindependent model reads the diff
03humanI read it, and I sign it
04gatetests, lint and build green
05shiprepo, runbook, handover
reviewevery diff, twice
you getthe repo and the runbook
lock-innone
4products live in public, all built solo
4open-source MCP servers on GitHub
177products measured by my agent-readiness scanner
10+years shipping software before the agents existed
What I build
Four kinds of work, and all of them end in production.
Every engagement starts from a product or a process that already exists, and ends with something you can observe, test and hand to your own team.
AI inside an existing product
Copilots, decision support, retrieval and generative features shaped around the way your users already work.
Product UXRAGTool useHuman review
MCP and agent-ready products
Secure MCP servers and agent interfaces that expose the right tools, data and workflows, with authorization, scopes, documentation and tests in the same delivery.
MCPOAuthTool designDiscoverability
Agent workflows and what they cost
Multi-step operational systems with approvals, recovery paths and audit trails, tuned against the cost of a finished task rather than the price of a token.
OrchestrationEvalsGuardrailsAI FinOps
Signature engagement
Agent-Ready Product Sprint
I take an existing SaaS or API and get it to the state where an agent can find it, understand it and use it. The sprint covers interface design, the MCP implementation, authentication, integration tests, machine-readable documentation and a readiness audit you can re-run yourself afterwards.
Ask about the sprint↗- 01Architecture
- 02Working implementation
- 03Agent test harness
- 04Launch plan
Live products
Four products, one engineer, all of them public.
This is the whole argument. One person, working this way, shipped all four. Open them and check.
AI product · MCP · finance
Agent Rynku↗
A production AI analyst for Warsaw Stock Exchange filings. One MCP data layer serves the in-product chat, a Telegram bot and any MCP client a user connects.
One MCP server exposing 92 tools over exchange data
Agent readiness · research
Let Agents In↗
A reproducible audit of whether an agent can find a product, read its docs, create an account, get a key and make the first API call without a person helping.
177 products measured, 79 with a signup an agent cannot render
Local AI · extension · MCP
Meet Live Assist↗
A local-first assistant that reads live meeting captions and answers in a side panel while the call is still running. The transcript stays a file on your own disk.
Chrome extension plus a local MCP server
Agent infrastructure · hosted
Muster↗
Shared operational memory for agents that outlive their own sessions: who is on duty, who owns what, what rotted and what needs a human. An agent registers itself with one HTTP call.
Signup is one curl, no account and no CAPTCHAHow the work runs
Agents do the volume. A senior engineer stays accountable.
Frame
We agree the outcome, the constraints and what evidence counts as done. This step is never handed to a model.
Run
Agents write the implementation, the migrations and the test suites. I decide what gets built, in what order, and what gets thrown away.
Review
Every change is read twice: by an independent model that never saw the reasoning, and then by me, because a model rationalises its own work.
Hand over
Observability, fallbacks and documentation ship with the feature. You end up with the repository, the tests and the runbook, not a dependency on me.
The parts vibe coding skips are the parts you are paying for.
- ✓Nothing ships that a human has not read
- ✓Tests that can actually fail, not tests that describe the code
- ✓You own the repository, the history and the runbook
- ✓A human approves anything with consequences
- ✓Deterministic code wrapped around probabilistic models
- ✓Cost, latency and failure telemetry from the first day
- Where
- Kraków, working remotely across the EU
- Languages
- Polish and English
- Experience
- 10+ years in product engineering
- Contact
- Straight to my inbox, no form and no funnel
The obvious objection
One person is the whole supplier. Here is why that is safe.
Everything lives in a repository you own, with the history, the tests and a runbook written for whoever comes next. No in-house framework, no hosting that only I understand, no month of archaeology if you replace me. Behind that sits more than ten years of full-stack product engineering, with the recent years spent on AI agents, MCP and production SaaS. When a scope genuinely needs a second specialist, I bring one in and tell you who it is.
Got a quote you are not happy with?
Let's find the smallest version of your project worth building.
Tell me the project, the deadline and what you were quoted. You get a technical answer from me, not a sales sequence.
Write to me↗Krystian Gwizdała · gwizdala.kr@gmail.com