Agentic engineering, by default
You hand me a goal, a few guidelines, or just a problem worth solving. I talk to the people who use the software, plan it, build it with coding agents under rules I enforce, verify it, ship it and measure it. You get working software in production, not moved tickets. Code, CI and docs stay in your repository and accounts from the first commit.
Calls this week · work starts 3 November 2026 · Wrocław, Poland (CET), remote
Sound familiar?
In METR’s randomized trial (10 July 2025), experienced open source developers took 19% longer to finish issues with AI tools, yet believed AI had sped them up by about 20%.
Seats are a tool, the workflow is the work. I run Claude Code and Codex by default under rules enforced by hooks, lint and test gates in CI, so the speed shows up as shipped releases: 270+ tagged releases of one product in 10 months of 2026.
Faros AI’s 2026 engineering report (telemetry from 22,000 developers, analysis as of March 2026) found median PR review time up 5x at high AI adoption, with 31% more PRs merged without any review.
I do. In 2026 I merged 260+ PRs of my own and reviewed 300+ of other engineers’. About two thirds of merged PRs across two products were built with coding agents, and nearly all went through my review or merge, behind gates in CI that catch what a tired reviewer misses.
Sound like you? Book a free 30‑min call →
Faros AI’s telemetry from 10,000+ developers (23 July 2025) showed teams with high AI adoption completing 21% more tasks and merging 98% more PRs, with no significant correlation to improvements at the company level.
I take a goal, not a ticket, and count what reaches users: 270+ tagged releases of one product in 10 months. I answered about 99% of Sandtime support myself, so I see what users hit after every release and fix that first.
Anthropic’s study of its own engineers (2 December 2025) found everyone becoming more full‑stack, and named a paradox of supervision: overseeing Claude takes the same coding skills that heavy AI use can erode.
15+ years of shipping is what lets me supervise agents and own the outcome. Give me a goal and I take it end to end: plan, build with agents, review, release and support. I led an app through Google’s CASA Tier 2 twice, passed in January and October 2026.
Facing one of these? Book a free 30‑min call →
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: 260+ PRs of my own and 300+ reviews of other engineers’. Lint, type and test gates in CI reject the mechanical problems before a human looks.
Where there is no product manager, I set priorities with the people who use the software and answer them myself: about 99% of Sandtime.io support went through me. Jira, Linear or GitHub Issues are fine if you have them; I use them when they help, and moving cards is not the job.
TypeScript, React and Node.js every day, ten years of C# and .NET, Python when the work needs it. Web, mobile, desktop, Cloudflare, Postgres, Docker, CI/CD, monitoring and logging: I own what I ship, after the deploy too. See the full‑stack page.
Done beats perfect, without shortcuts where they are risky. 270+ tagged releases of one product in ten months of 2026, every change behind tests, and every feature measured in PostHog. A product I led passed Google’s CASA Tier 2 security assessment in January and again in October 2026. Outside checks agree: the product I led passed Google’s CASA Tier 2 security assessment in January 2026 and again in October 2026. How the agents work under rules is on the AI consultant page.
Live demo
A replay of a real feature on Sandtime.io, built at Sanddev: people could not connect Claude’s web connector because it has no field for an API key. Step through it as it happened: the goal, the plan, the agent session, the gates that failed, the hardening, the review and what shipped.
Every step is from the actual pull request, with names removed. Bring your own goal to the call and we sketch its run the same way.
Ways to work together
Want a number first? The configurator shows hours, a price range and the start date. The call is free either way.
goal
One outcome and a budget: “customers can pay by card”, “the import stops failing”, “agents can use our API”. I come back with a plan within days and ship the first slice in week one.
Book a free call →guidelines
Your architecture, your conventions, your review. I work inside your repositories and CI, write the rules your agents follow, and keep to your definition of done.
Book a free call →just act
A spending cap and a weekly call. I pick the highest-impact work from your users, analytics and backlog, ship it, and show you the numbers every week.
Book a free call →full capacity
About 167 hours a month of senior work for $7,500 net, three months minimum, one month’s notice on either side. The best value when you have months of work.
Book a free call →Proof
Problem, what I did, result. Client names only where I am allowed to use them.
See also: AI & agents · AI in your company
Facing something similar? Book a free call →
How it works
You describe the product and the problem. I ask the questions I would ask on day one.
Within 24 hours after the call you get a short written plan: what we build first, what we measure, how we work together and what it costs. No slide deck.
Deployed, measured and open for your review. After that, the numbers set the pace.
Fit check
Working style
Claude Code and Codex work on main with hooks, lint and test gates. The model suggests, the criteria decide. Every PR gets a human review and AI‑assisted content is disclosed.
Where there is no PM I set priorities directly with the people using the software and answer support myself.
A bug is fixed or consciously closed, never parked. Alerts and error tracking in every project, so I learn about problems before the client does.
Good to know
METR’s 2025 trial found experienced developers 19% slower with AI tools. On 24 February 2026 it changed the study design because many developers no longer agreed to work without AI, and it called the new data a bad proxy for real productivity (METR).
In the 2026 Stack Overflow survey (6 October 2026), 66% of coding agent users used Claude Code and 59% GitHub Copilot, and 48% of developers trust AI output when they can easily validate it (Stack Overflow 2026).
Anthropic puts average enterprise Claude Code cost at about $13 per developer per active day and $150 to $250 a month, below $30 a day for 90% of users. Agent teams use about 7 times more tokens than a standard session (Claude Code costs).
Locally, Codex runs in an OS‑enforced sandbox with network access off and writes limited to the workspace. Turning on network or live web search widens prompt injection exposure (Codex security).
A claude -p session, the usual mode in CI, shows neither the workspace trust dialog nor the approval prompt for servers in a repo’s .mcp.json. Anthropic also does not security-audit MCP servers (Claude Code security).
On 26 May 2025 Invariant Labs showed a prompt injection in a public issue making an agent with the GitHub MCP server copy private repository data into a public pull request (Invariant Labs).
FAQ
Bonus · #suchar
Free, like the first call. Same sense of humor, included in every project.
It ran out of context. Ba‑dum‑tss.
“Great question! You’re absolutely right.” Ba‑dum‑tss.
A Claude Monet. Priced per token. Ba‑dum‑tss.
You bring the problem, I bring the questions. Want a number first? The configurator shows hours, a price range and the start date.
or write to ciunkos@gmail.com
After you book you get a calendar invite. Nothing else to fill in.
My pledge
Calls this week · work starts 3 November 2026 · B2B contract · Wrocław, Poland (CET), remote
