Agentic Workshop
Book discoveryFor engineering teams · Zürich · EN / IT / ES
Ship with a shared agent method, in one half-day.
Live workflow on your backlog under NDA. Playbook, checklist, pilot backlog. Lite from CHF 2’500 · half-day CHF 4’500–5’500.
Tool-agnostic. Bring what you use
Patrick Barattin
Zürich · EN / IT / ES
Founder, Weft Labs · ex-CTO · Conversational AI at On · builds agent systems daily
How we work
Official agent tooling + a method you keep: backlog → session → review → playbook.
Ways to work
Lite, half-day, or a fixed consulting block after.
Workshop Lite
From CHF 2’500 · ~2h
Smaller groups, a first look, or when you need the method without a full half-day.
Half-day
CHF 4’500–5’500 · ~3–3.5h
The main offer: live backlog work under NDA plus full artefacts and pilot plan.
Main offer
Consulting hours after
Scoped separately · Time-boxed
A fixed block to refine the playbook and unblock the 30-day pilot, then done.
Included artefacts
Live session on your backlog
Under NDA: one real issue → plan → tested change on your stack, not a toy demo.
Team playbook
How your team runs agent-owned work: roles, handoffs, and when a human must decide.
Review checklist
What “good” looks like for agent-assisted PRs, so review stays consistent.
Repo instruction starter
A first cut of project instructions / agent context for your repos (you keep and edit it).
30-day pilot backlog
A short, ranked list of workflows to try next, with owners and success criteria.
After the method · optional OSS hosts
Create your own software factory
The half-day sells a shared method and artefacts you keep, not another agent IDE. Once that leave-behind exists, these open-source stacks are honest starting points for a coding factory around your repos (Cursor, Claude Code, Copilot, Orca, or what you already run). Use the checklist in discovery; we wire one stack into your pilot, we do not run a free tools tour.
What a factory needs
Tick the capabilities you need after the workshop method is in place. Cards re-rank so you compare OSS hosts, not shop for another IDE trial.
Open-source starting points
Curated GitHub repos. Stars are approximate snapshots (Oct 2026), not live. Not a vendor pitch: the workshop sells the method and leave-behind; these are optional hosts once that method exists.
Mastra Software Factory
6/8 fitTemplate for an agent-powered delivery board: GitHub / Linear intake to coding agents to PRs you host. Closest literal “software factory” starter.
No SPDX yet · ~50+ ★ · Active · GitHub ↗
OpenHands
6/8 fitSelf-hosted control center: sandbox runs, automations, issue-to-PR workflows. Strong when you want a factory host, not only an IDE.
MIT · ~90k+ ★ · Active · GitHub ↗
Orca
6/8 fitADE for parallel agents: worktrees, Claude Code / Codex / OpenCode side by side, GitHub and Linear in-app. Closest to a multi-agent factory desk.
MIT · ~87k+ ★ · Active · GitHub ↗
Aider
5/8 fitTerminal pair programmer with git-native commits and repo mapping. Fast supervised cell; you still own the review bar.
Apache-2.0 · ~49k+ ★ · Active · GitHub ↗
Cline
5/8 fitCoding agent as VS Code extension, CLI, or SDK. Best as a supervised cell inside a factory, not the whole factory.
Apache-2.0 · ~70k+ ★ · Active · GitHub ↗
SWE-agent
5/8 fitIssue-to-fix harness with a constrained agent computer. Useful backlog intake pattern; method and review still yours.
MIT · ~20k+ ★ · Active · GitHub ↗
Method first, then a stack. On discovery we map one of these (or what you already run) into the half-day leave-behind. You do not pay for a tool tour.
Book discoveryBefore you start
NDA
We sign first so we can work on a real backlog item, not a toy repo.
One real backlog issue
Pick a concrete ticket your team actually cares about. That’s the live workflow.
Tools
Bring Cursor, Claude, Copilot, or whatever you already use. Tool-agnostic by design.
Who attends
Best at about 6–12: leads and ICs who will own the pilot, not a spectator room.
Fit check
Good fit
- ✓ Engineering leads and senior ICs who want a shared method, not another tool trial
- ✓ Teams of roughly 6–12 who can join one half-day remote (or on-site in Switzerland)
- ✓ Shops already using Cursor, Claude, Copilot, or similar (tool-agnostic by design)
Not a fit
- – Teams looking for a vendor pitch or a product to buy
- – One-off “AI magic” demos with no artefacts to keep
- – Groups larger than ~12 in a single session (we’ll split or scope Workshop Lite)
Who runs it
Founder, Weft Labs
Zürich, 2023–present. Product and engineering for a platform that lets AI agents reach external data, software, and services through one integration.
Conversational AI at On
From Aug 2025 on the Conversational AI team: agent architecture, monitoring, and live evaluation for an internal ecommerce platform.
ex-CTO, Pix Group
Led 5 engineers and product experts while digitizing personalized apparel; scaled ops and platforms end to end.
Full detail on the CV.
FAQ
- What does it cost?
- Workshop Lite (≈2h) from CHF 2’500. A full half-day is typically CHF 4’500–5’500 depending on team size, remote vs on-site, and prep. A fixed consulting block after the workshop is scoped separately. You’ll get a written scope before anything is booked.
- How many people?
- Best at about 6–12. Larger groups work as Workshop Lite or split sessions. We’ll decide on the discovery call.
- Which tools (Cursor, Claude, Copilot…)?
- Tool-agnostic. We work with what you already use. The method (issue → plan → tested change, review bar, repo instructions) transfers across IDEs and agents.
- NDA?
- Yes. We work under NDA so we can use a real backlog item on your stack.
- Remote or on-site?
- Remote by default. On-site across Switzerland when that helps the team move faster.
- Languages?
- English, Italian, or Spanish. Pick what works for the room.
- What does “after” support mean?
- A fixed consulting block for the 30-day pilot: refine the playbook, unblock early PRs, then done. Time-boxed and scoped up front.
- How fast do you reply?
- I aim to reply within one business day. Book discovery on the calendar for the fastest path.
- Can’t we just install OpenHands, Orca, or Cline ourselves?
- Yes, and many teams should. The workshop is not another agent IDE. You leave with a shared method, review bar, repo instructions, and a 30-day pilot on your backlog. Tools are inputs; the leave-behind is what makes agent work stick.
- What is a software factory in this context?
- A repeatable loop: backlog intake → agent session → human review → PR leave-behind, with repo instructions so the next run is not ad hoc. The half-day installs that method. OSS repos below are optional starting points after you have the method.
Patrick Barattin · Founder, Weft Labs · ex-CTO · Conversational AI at On
Free 30-min discovery
Written scope before booking · patrick@weftlabs.com
Lite from CHF 2’500·Half-day CHF 4’500–5’500EN / IT / ES
Book discovery