Przemysław Zalewski

Coding your own system with AI?

A senior next to you, while you build your own system

You are a programmer out of necessity: you run a transport firm, a workshop or an agency and built the system it runs on with an AI agent. Some questions the agent will not answer well: how to keep many clients’ data apart, where backups live, why it loops on the same error. You keep building; I sit next to you 8 to 16 hours a month.

See price and start date → Book a free 30‑min call

8 to 16 hours a month · $36⁠–⁠54/h excl. VAT Details in the configurator →

Przemysław Zalewski

Sound familiar?

Problems owners who code with AI bring me in 2026

Can Claude Code or Cursor delete my production database?

In April 2026 a Cursor agent running Claude Opus 4.6 deleted PocketOS’s production database and its backups in nine seconds, using credentials it found while working in staging, with no confirmation step; the newest recoverable backup was three months old until the hosting provider later recovered newer data. An agent holding production credentials can do whatever those credentials allow.

We separate development and production on your accounts, keep production keys out of the agent’s reach and set up backups, then restore one together so you know it works. I build guardrails into the environment, not into people’s memory: I once set up staging so it physically could not email anyone outside our own domains.

Is the npm package my AI agent installed safe?

On 20 April 2026 CISA warned that axios, one of the most used npm packages, had two malicious versions published on 31 March 2026 that pulled in a dependency downloading a remote access trojan, and told teams to rotate npm, CI/CD, cloud and SSH secrets. Agents add dependencies in seconds, and nobody reads the lockfile.

We pin and review dependencies together, add lockfile checks to CI and upgrade in waves you can revert, a typical app often in one evening. In October 2026 I configured an app’s npm installs to take only versions published at least seven days earlier, so a hijacked release is usually caught before it reaches us, and kept coding agents from installing packages without asking.

Did my AI coding agent leak my API keys to GitHub?

GitGuardian’s 2026 report counted 28.6 million new secrets in public GitHub commits in 2025, and commits co-authored by Claude Code leaked secrets at about twice the baseline rate. A leaked key works for whoever finds it until you rotate it.

We move keys into your platform’s secret store, turn on secret scanning before every push and write the rule into your CLAUDE.md or AGENTS.md, so the agent stops putting keys in code. In October 2026 I added a secret scan to every push and pull request and took production keys out of CI.

Facing one of these? Book a free 30‑min call →

Scope

What we do together

Decisions that are hard to undo

Data model, many clients in one database, which hosting, which payment provider. Decided with you and written down in your repository.

Deploys, backups and environments

Set up together on your accounts, so a change goes out from the repository and a backup really restores.

Getting unstuck

When the agent loops on the same error, we find the cause on your screen instead of another prompt.

Rules for your agent

A CLAUDE.md or AGENTS.md written for your project, so the agent stops repeating the same mistakes.

Process

How it works

  1. A first call

    What you build, what blocks you, which decisions are coming.

  2. 8 to 16 hours a month

    Live sessions on your screen and questions on chat between them.

  3. You keep the code

    Everything stays in your repository and on your accounts. I do not take over unless you ask.

  4. When it grows

    If clients arrive, the same person can help turn it into a product or join for longer.

Who it is for

  • Owners who build their company system themselves with an AI agent.
  • Founders whose MVP from Lovable or Bolt now has its first clients.
  • Small teams without a senior who want one a few hours a month.

Track record

Proof

The first case is this offer in practice: rules in the repository that keep a coding agent in line. The second is the same setup restarting delivery on a mature SaaS, the third four side projects of mine on one domain.

An AI coding agent as a team member, under enforced rules

The problem
Coding agents write code faster than a team can review it. A team of four wanted the speed without a slop swamp on main.
What I did
I authored the rule set shared by Claude Code and Codex: mechanical checks over written rules, lint and test gates for security, API surface and CSP, agents working on main with hooks, every PR reviewed by a human, AI‑assisted content disclosed as such.
Result
In 2026 about two thirds of merged pull requests across two products were built with coding agents, and nearly all went through my review or merge. I merged 260+ PRs of my own and reviewed 300+ of other engineers’. My commits across those two products went from under 1,000 in 2025 to 7,500+ by October 2026.

~2/3 PRs with agents 300+ reviews 7,500+ commits in 2026

See also: AI & agents · Full‑stack & custom software

Offer: AI engineer for your team (Node.js, LLM agents)

A new engineer from day one to owning a product

The problem
A new engineer joined with zero context: new codebase, new domain, first week.
What I did
Pair-coding sessions from day one, twice a day in the first week, a working build configuration and a written release procedure instead of “read the docs”, then a year of shared work under the same agent and review rules. I onboarded two engineers that week.
Result
By day five she proposed her own component. Within a year she committed about as much as I did. She now reviews senior pull requests and owns a product that had one lead for eight years, and both engineers kept the agent-review practice after I left.

day 5 1 year owns a product

See also: Full‑stack & custom software

Four side projects on Cloudflare Workers under one domain

The problem
Ideas die when deploying them costs a weekend. I wanted a path from idea to a URL in an evening.
What I did
Korpo Tłumacz, Wyrocznia, Migawki and Przemo (Agent): each a Worker with static assets, Workers AI with per‑IP and global rate limits, a cost cap in a Durable Object, R2 for generated images, a first-party PostHog proxy and a CSP without inline scripts.
Result
Live at ciunkos.com/korpo-tlumacz, /wyrocznia, /migawki and /przemo. This page runs on the same pattern.

Workers AI Durable Objects R2 Open ↗

See also: Cloudflare · AI & agents

Offer: Company login for apps your people built with AI

Want the number before the call?

8 to 16 hours a month · $36⁠–⁠54/h excl. VAT

See price and start date → Book a free 30‑min call

Good to know

Things that bite owners who code with AI

/rewind does not undo shell commands

Claude Code checkpoints track its own file edits only. Files changed by Bash commands, such as a migration script, stay as they are; only git brings them back (checkpointing).

Checkpoints expire after about 30 days

Checkpoint snapshots are deleted about 30 days after the session last saved one (checkpointing).

CLAUDE.local.md switches off AGENTS.md

Adding a personal CLAUDE.local.md in a project that relies on AGENTS.md stops Claude Code from reading AGENTS.md (Claude Code docs).

Supabase backups skip uploaded files

Database backups do not include objects in Supabase Storage, so uploaded files need their own copy (Supabase backups).

Stripe test mode hides retry bugs

Live webhooks retry for up to 3 days, sandbox ones only 3 times over a few hours. Events can also arrive out of order or twice (Stripe webhooks).

Webhook signatures need the raw body

Stripe signature checks fail if the framework parses the body first, a classic bug in generated webhook code (Stripe webhooks).

FAQ

Questions

Will you check or grade my code?
No. We work on what you are building next, together, and the decisions stay yours.
I am not a programmer. Is this for me?
Yes, if you already build with an AI agent and want to understand what it does. We talk in plain words.
Can I change the hours later?
Yes, between 8 and 16 hours a month, from one month to the next, with no minimum term. Unused hours do not roll over.
What does it cost?
8 to 16 hours a month at $36 to $54 an hour, excl. VAT, depending on how hard the work is. The configurator shows the price and the start date before we talk.
What if I want you to build something?
Then it becomes a separate project with its own price, for example turning your tool into a SaaS.
Claude Code broke my database. Can /rewind undo it?
Only for files Claude edited with its own tools. Shell commands and migrations are not tracked, so commit often and keep real backups (checkpointing, Supabase backups).
Why does my Stripe webhook work in test mode but fail live?
Common causes: a parsed body breaks the signature check, events arrive out of order or twice, and test mode retries far less than live (Stripe webhooks).
Can Claude Code or Cursor delete my production database?
In April 2026 a Cursor agent running Claude Opus 4.6 deleted PocketOS’s production database and its backups in nine seconds, using credentials it found while working in staging, with no confirmation step; the newest recoverable backup was three months old until the hosting provider later recovered newer data. An agent holding production credentials can do whatever those credentials allow. We separate development and production on your accounts, keep production keys out of the agent’s reach and set up backups, then restore one together so you know it works. I build guardrails into the environment, not into people’s memory: I once set up staging so it physically could not email anyone outside our own domains.

In 30 minutes you will know if we are a fit

You bring the problem, I bring the questions. Want a number first? The configurator shows hours, a price range and the start date.

Book a free 30‑min call Email me instead

or write to ciunkos@gmail.com

After you book you get a calendar invite. Nothing else to fill in.

My pledge

  • Reply within 24 hours. I reply within 24 hours on working days, personally, not a bot. Weekend messages: by Monday noon.
  • Late means free. If I am late, the first hour of work is free. You do not have to ask.
  • Scope in writing within 24 hours. A free 30‑minute call with no company deck, then the scope and the price in writing within 24 hours. You decide.
  • The person who does the work. You talk to the person who does the work. No sales layer, no handoff.
  • Follow‑up only with news. I follow up only when I have something new for you. Never a “did you see my offer”.
  • Honest “not a fit”. If I am not the right person, I say so on the call and point you to someone who is.

Calls this week · work can start now · B2B contract · Wrocław, Poland (CET), remote

You own the code, domain and accounts from day one. Nothing is held hostage.