Przemysław Zalewski

AI and Forward Deployed Engineer, contract or full time

An AI engineer who has shipped agents to production

Node.js and TypeScript for nine years, LLM agents in production since 2024. I build the tools, the evaluation loop and the cost cap, and I keep coding agents under rules enforced in CI. Contract or full time, available now.

Book a free 30‑min intro call →

Dedicated contract: $7,500 a month, or hourly at $36⁠–⁠54/h excl. VAT. Details in the configurator →

Przemysław Zalewski

Sound familiar?

Problems teams hire an AI engineer for in 2026

Who maintains the AI agents a vendor built for us?

Gartner predicts that by 2028, 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering, trapped by soaring costs and unable to evolve it on their own (29 September 2026).

I work inside your repositories, on your stack and your review process, and every piece ships with tests and a runbook your team can run. You own the code, and the knowledge is written down in your repo. In September 2026 an AI crawler I maintained had kept its server at 100% CPU for weeks with the disk 96% full; I capped it at three browsers with hard deadlines and memory limits and cleaned up about 1,200 abandoned browser profiles per container.

Is code from AI coding agents secure?

Veracode’s 2026 GenAI Code Security Report, published on 28 July 2026, found that AI‑generated code passes security checks only 56% of the time on average, virtually unchanged since 2025. Even the best model fails nearly one in three tasks, and coding-specialized models are no safer than general ones.

Agents work under rules enforced by hooks, lint and test gates in CI, and a human reviews before merge. On an AI email client I turned the CASA and OWASP ASVS controls into Playwright tests on every pull request, and led Google’s CASA assessment to a pass twice, in January and October 2026.

Sound like you? Book a free 30‑min call →

Can a malicious npm package hijack our AI coding tools?

On 4 August 2026 Snyk reported 11 malicious releases of keyv and related npm packages. Besides an install script, they planted a .claude/settings.json whose SessionStart hook runs the attackers’ loader once a developer or coding agent trusts the project’s configuration.

I upgrade dependencies in reviewed waves you can revert, a typical app often in one evening, with lockfiles pinned and agents on narrow permissions. When React2Shell hit in December 2025, my clients’ apps were patched the same day.

Can a cloned repo take over a developer’s Claude Code?

CVE‑2026‑55607, published 25 June 2026, CVSS 7.7: in Claude Code from 2.1.38 before 2.1.163, a cloned repository with prompt injection content could confuse git worktree handling, write files such as ~/.zshenv in the user’s home directory and run code outside the sandbox.

Agent settings, hooks and MCP servers live in the repo and go through the same review as code, and the agent tools get updated in the same waves as every other dependency. In October 2026 I pinned a repo’s Claude Code start hook to a vetted package copy, added a test that blocks hooks fetching packages at run time, and kept production secrets out of agent sessions.

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

Scope

What I bring to your team

Agents in production

Tools with narrow scopes, an evaluation loop against fixtures, a cost cap, models compared between releases. No framework you cannot debug.

MCP servers and integrations

Your product usable from ChatGPT and Claude, on top of your existing login, with adoption measured per client.

Coding agents under rules

CLAUDE.md and AGENTS.md, skills, hooks and CI gates. In 2026 about two thirds of merged PRs across two products were built with coding agents, and nearly all went through my review or merge.

The rest of the stack

React and Next.js, Node.js, PostgreSQL, Cloudflare Workers, PostHog. I ship the whole feature, from the database to the screen.

Process

Terms

  1. Dedicated contract (B2B)

    Full capacity for 7,500 USD, 6,700 EUR or 30,000 PLN net a month, about 167 hours, three months minimum. Or hourly at 36 to 54 USD. Invoiced from Poland in the currency you prefer.

  2. Employment

    Open to a full‑time role as an AI engineer or AI enablement lead. Salary to agree on the call.

  3. Start

    Now. Remote from Wrocław (CET): the same hours as Berlin and Paris, about three shared hours with the US East Coast.

  4. Process

    A 30‑minute call, a technical conversation with your team, and instead of homework a paid two‑week trial: five tasks from your backlog, 2,880 to 4,320 USD excl. VAT, price fixed before we start.

Setting up a team of AI builders?

  • If your company is starting a department where people build tools with Claude Code, Codex and n8n, I can set it up and run it: shared rules, review, deployment, costs.
  • Your people keep building; I keep the quality bar and the bills under control.

Forward Deployed Engineer

An engineer who sits with the customer or the business team and turns a model into a system that works there.

With the people who use it

I work next to the team that has the problem, in their tools and their data, and ship in small steps they can try the same week.

Integrations first

MCP servers, APIs, ERPs, mailboxes, document flows: most of the work is connecting the model to what the company already runs.

Measured in production

Evals on real cases, cost caps and usage per person, so the decision to keep or drop a feature is made on numbers.

Handed over, not owned forever

Rules, runbooks and a team that can run it, whether I stay on or move to the next deployment.

Track record

Proof

15+ years of shipping software. Three recent AI cases below, each with its numbers. Products built at Sanddev are named and live, so you can check them; client names stay private.

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

An MCP server from first merge to first external users

The problem
Users of Sandtime.io, the time tracker I built at Sanddev, used by 4,000+ freelancers and teams, wanted to log time from their AI assistants. The API existed, the agent surface did not.
What I did
I set the tool surface, removed destructive tools, routed every operation through the existing REST controllers so an agent gets no wider access than a person, and instrumented adoption per client in PostHog from day one.
Result
First external users. A bot-protection rule that rejected MCP clients by user agent was found and removed the same day because the dashboard showed it.

MCP OAuth adoption dashboard

See also: AI & agents · PostHog & product analytics

Offer: Your company systems inside Claude and ChatGPT (MCP)

Patriot Mail: an AI email client through CASA Tier 2 and onto the App Store

The problem
A greenfield AI‑native email client for web, iOS and macOS, built at Sanddev, needed Google’s CASA Tier 2 security assessment before it could read Gmail, and it had to work for older people.
What I did
I was the principal engineer at Sanddev. I wrote the delivery plan, shaped the product vision and treated prompt injection through untrusted email as a design constraint before any model touched a mailbox. CASA and OWASP ASVS controls became a Playwright suite that runs on every pull request. WCAG 2.2 AA sits behind an automated axe gate.
Result
I led CASA Tier 2 to a pass in January 2026, with third-party pentest findings fixed, and led the yearly reassessment, validated in October 2026; the questionnaire turned into a check on every PR. Live on the App Store on iOS and macOS since April 2026. Text scaling to 200% and a calm‑email mode shipped for older users. At that reassessment five items came back for missing evidence; I fixed the code first, then resubmitted updated evidence, and all five were accepted on 27 October.

CASA Tier 2 iOS + macOS WCAG 2.2 AA Open ↗

See also: SaaS & MVP · Mobile & desktop apps

Offer: From an AI‑built MVP to a product people pay for

Want the number before the call?

Dedicated contract: $7,500 a month, or hourly at $36⁠–⁠54/h excl. VAT.

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

Good to know

Production LLM details an AI engineer should know

Long context costs the same per token

On Claude 4.6 and later, except Haiku 5.5, a 900k‑token request is billed at the same per‑token rate as a 9k one (pricing).

US-only inference costs 10 percent more

Pinning Claude inference to the US multiplies every token price by 1.1, cache reads and writes included (pricing).

Batches can pass the spend limit

Claude batches may go slightly over the workspace spend limit, so cap batch size on your side (batch processing).

Expired OpenAI batches are still billed

When an OpenAI batch expires after 24 hours, completed requests inside it are still charged (OpenAI batch).

Four cache breakpoints, 20 blocks back

Claude allows 4 cache breakpoints, and each looks back only 20 blocks, so long tool loops can miss the cache (prompt caching).

Only exit code 2 blocks a hook

In Claude Code, a PreToolUse hook blocks a call only with exit code 2; other non‑zero codes let the action go ahead (hooks).

FAQ

Questions

Are you available for a full‑time role?
Yes, starting now, as an AI engineer or AI enablement lead. B2B contracts work too.
What is your rate?
A dedicated B2B contract at full capacity is 7,500 USD, 6,700 EUR or 30,000 PLN net a month, three months minimum. Hourly, 45 USD, 40 EUR or 180 PLN at the base, 20% less for simple work and 20% more for complex work. Pricing has both. For an employment role, we agree on the call. For comparison, Arc.dev puts senior freelancers in the US at 95 to 150 USD an hour (May 2026, costs compared).
Which models do you work with?
Whatever wins the evaluation for the task: Claude, OpenAI models and open models, compared on the same fixtures.
Can I see your CV?
Yes, at ciunkos.com/resume.
What if it does not work out?
Test me first with the two‑week trial from the terms above: five tasks from your backlog, price fixed before we start, code in your repo. Hourly work has no minimum. The dedicated contract runs three months minimum, then one month’s notice on either side with a written handover.
You are one person. What if you get sick or leave?
The work does not leave with me. Code, CI and docs live in your repository and accounts from day one, every project keeps a runbook and architecture notes written from the code, and notice is one month on both sides with a written handover. If I am ill, you hear it the same day and the plan moves, not the quality.
Who owns the code, and who sees our data?
You own the code. It lives in your repository and accounts from the first commit, and the contract transfers the copyright to you in writing. Only I work on it, with no subcontractors. Coding agents run with training off, and secrets, customer data and logs never reach a model without your written approval. Facts for your vendor questionnaire are in the trust center.
Is a 1M‑token context more expensive per token on Claude?
Not on Claude 4.6 and later, except Haiku 5.5. US‑only inference adds 10 percent to every token category (pricing).
Can the Batch API blow my budget?
Claude batches can slightly exceed a workspace spend limit (batch processing), and OpenAI bills completed requests even in an expired batch (OpenAI batch). Cap batches in your own code.
Who maintains the AI agents a vendor built for us?
Gartner predicts that by 2028, 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering, trapped by soaring costs and unable to evolve it on their own (29 September 2026). I work inside your repositories, on your stack and your review process, and every piece ships with tests and a runbook your team can run. You own the code, and the knowledge is written down in your repo. In September 2026 an AI crawler I maintained had kept its server at 100% CPU for weeks with the disk 96% full; I capped it at three browsers with hard deadlines and memory limits and cleaned up about 1,200 abandoned browser profiles per container.

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.