
Agentic Workflow Sprint
In 2–4 weeks we design and build repeatable AI workflows or an agent prototype running on a real process.
- Workflow spec and prototype
- Human-or-AI approval modes
- Before/after metrics and guardrails
- Decision: scale or kill

In 1–2 weeks we map your products/services, processes and data, then pick the AI use cases with the highest odds of a measurable effect. The result is a decision and a backlog of first experiments.
For companies that know AI matters but haven’t picked the process to change yet.
In 1–2 weeks I go through every place you use AI: whether it’s the best setup of models and tools for that task today, and where AI does a fraction of what it could. The result is a list of upgrades with the expected effect next to each.
For companies that already run AI in their product, processes or team licences and want to know how much of it actually works for them.
In 2–3 weeks I work in your code and bring the cost per task down without losing quality. The first three days measure an eval set built from your real traffic. Then every change passes a quality gate, ships behind a flag and can be rolled back.
For companies with LLMs in production that pay a real API bill every month and watch it grow.
One department from your company. In 2–4 weeks I test whether AI can automate its work — fully or partially — with an owner, data, human-or-AI approval modes and before/after metrics.
For founders, CTOs and product teams that want real change in delivery, support, sales ops or product.
The agent fails in production, the model hallucinates, retrieval misses — and the team keeps going in circles. In 1–2 weeks I find the cause and clear the blocker, with before/after metrics.
For teams that already built something and hit a wall — the problem has a name, progress has stopped.
I set up how your team works with coding agents: from ticket, through code and review, to shipping. Ticket patterns, repo instructions and a pilot on your team’s real tasks — measured.
For product-engineering teams that have the AI tools but use them chaotically.
Regular sparring for people making decisions about AI, product and technology: roadmap and architecture review, tool and vendor assessment, technical and product priorities.
For founders, CTOs and product leaders — monthly billing, leave anytime.
They have a problem getting from experiments to change in day-to-day work. People use ChatGPT or Claude. Teams test Claude Code, Claude Cowork or ChatGPT Codex/Work (the optimistic scenario) or Copilot (the pessimistic one). The board has seen a few demos — everyone was impressed. But daily work looks exactly like before.
AI only starts to matter when it changes one of four things:
from ticket to shipped faster than the team alone
the same work done cheaper, with less manual effort
better data, faster research, less guessing
a small team handling volume that used to need an entire department
A few weeks after the Sprint ends, you have on the table:

You’re not behind — almost every company’s AI is exactly where yours is: at the demo stage. I build with AI daily, and most experiments don’t survive. The difference is that I measure them and kill the losers.
I’ve been building digital products since 1998. I co-created Nokaut.pl and Morizon.pl — from startup to IPO and exit. Today I build AI products and work with agents daily: an agent takes a ticket, writes code, another model reviews it, a human approves the final result. Not a futuristic deck — my regular workday.

In 2–4 weeks we design and build repeatable AI workflows or an agent prototype running on a real process.

In 1–2 weeks we map your products/services, processes and data, then pick the AI use cases with the highest odds of a measurable effect.

I set up how your team works with coding agents: from ticket, through code and review, to shipping.

Regular strategic sparring for founders, CTOs and product leaders making decisions about AI, product and technology.

I check how much of the AI you already pay for actually works for you — and which setup of models, tools and vendors fits each use case best.

I bring the LLM bill of a production system down without losing quality: changes in your code, each behind a quality gate on an eval set from your real traffic.

I solve the hard AI problems your team can’t crack — in your codebase, with before/after metrics.

In 5 days I test a prototype that stalled after the demo and give you a verdict: ready for a Sprint, fix first, or kill it.

Natural-language search and recommendations on your catalogue, compared with your current search on real queries.
It’s for companies with a process worth changing, an owner for it, and the readiness to decide based on the numbers.
I assess the AI in companies before you invest, and help cohort or portfolio teams measure whether their AI works. The fund or programme organiser pays.

In 3–10 days I check how much of a company’s “AI edge” holds up: code, data, model costs and team.

Two sessions for 8–15 teams. Each leaves with a number it measures and the date of its next decision.
No. I can train your team if needed, but the goal isn’t tool knowledge. The goal is changing a specific process, product or way of working.
Depends on scope. Most often I deliver a prototype, a workflow spec, a playbook and an implementation recommendation. I can build the production version with trusted partners, or — on bigger projects — join as an advisor working with your team. I build production-grade software with AI every day.
That’s also a result you’re paying for: a “kill” verdict with the data on why — instead of quarters spent dragging a pointless project. The Sprint is capped at 2–4 weeks, doesn’t slide into an open-ended engagement, and the deliverable isn’t a deck — it’s a working workflow and its numbers.
Every service page shows a starting price, net. The Stalled AI Prototype Review starts at €2,200, and if you order a Sprint within 30 days its full fee comes off the Sprint price. A Sprint starts at €9,500, depending on integrations and data sources. Advisory starts at €1,200 a month. The exact fee is fixed in writing before work starts: message me on LinkedIn or email michal@jaskol.ski with the process in question, and I’ll reply within 1–2 business days.
Sometimes, but not immediately and not everywhere. First we define the approval mode: outputs are approved by a human or — where it’s safe — by AI. The bigger the blast radius, the more control, logs and human approval.
Yes — if the project has an owner, a clear problem and room for a fast test. I’m a poor fit for a six-month slide process with no decision.
Usually with an AI Opportunity Map or an Agentic Workflow Sprint. The first gives you a decision, the second a working prototype. If you already have a prototype that has stalled for months, start with the Stalled AI Prototype Review: five days and a verdict.
Yes. For funds I run AI due diligence: in 3–10 days I check a company’s code, data, model costs and team, and the memo ends with a recommendation. For programmes and funds I run an AI clinic where every cohort or portfolio team leaves with a measured number and the date of its next decision. In both cases the fund or organiser pays.
Ten yes/no questions. Two minutes. An honest answer: sprint, opportunity map — or not yet.
Check if your process is Sprint-readyThe shortest path: a message on LinkedIn. We’ll check whether your process fits a Sprint — and whether it’s worth it at all.
Or just email me: michal@jaskol.ski
Prefer the form? Describe the area/process/challenge — I reply within 1–2 business days.