00Barcelona, working everywhere
Jasper Staats, partner and Head of Tech at Like A Human. I build the platform behind every workshop and the agentic tooling we teach.
01Who I am
Ten years as a lead engineer taught me how software gets built. AI changed how fast. Now I direct a crew of AI agents, build the platform behind every Like A Human workshop, and stand in front of rooms full of engineers, marketers and designers to show them how it works in their own tools.
Jasper Staats
Co-runs Like A Human with Stefan Luttik and Tim Weeda.
01.1Where the ten years went
Partner and Head of Tech
The platform behind every workshop, the agentic tooling we teach, and a place in front of the room.
02On stage
Speaker at official Anthropic Claude Code events, hackathons and masterclasses in Barcelona and Amsterdam. When I can, I skip the slides and build in front of the room.
Claude Community Event, Barcelona, April 2026. Photo: French Tech Barcelona
Claude Community Event, Barcelona, 9 April 2026
Anthropic invited Like A Human on stage to talk about something almost nobody covers yet: creativity inside Claude Code. I ran the eight tentacles live, my MCP integrations, agents and custom creative skills, and let the audience pick the subject. A full website with video, made from scratch in real time.
The applause came during the demo, and a line of people afterwards wanted to know how.
See the post from the day
What I talk about
Beyond prompt engineering
Context engineering in practice: project memory, skills, agents and how to keep a long session sharp.
The context window
What a model actually sees, why it forgets, and how to design work around it.
How these models work
From a generative pre-trained transformer to the tools a team uses on Monday.
03What we teach
Like A Human runs hands-on Claude workshops for engineering, business and creative teams. No slide decks. Everyone works in their own tools, on their own work, until it sticks.
CDA01 of 03
For engineering teams
Developers, tech leads and product owners learn to plan, build and ship with Claude Code. Not autocomplete: agents that take a ticket, work in their own worktree and come back with a change that passes the checks.
The same agent loop I run on our own platform every day, taught on your codebase.
CBA02 of 03
For business teams
Marketing, sales, operations, finance and HR map their own workflows and hand the repetitive parts to Claude Cowork. Everyone leaves the day with a workflow that runs in their own tools.
Guided build: Grand Café Lumière. I build the platform the team works in, from the first lesson to the last check.
CCA03 of 03
For creative teams
Designers, art directors and content teams pair Claude Cowork with Krea to go from a brief to an image campaign that reads as photography, not as AI.
Guided build: Studio Lumière, an image campaign made with Cowork and Krea. I bring the image method I use on my own work.
04Way of working
I build platforms the way I would want to inherit them: one source of truth, clear rules, and room for agents to do real work. It is how the platform behind Like A Human is built. Pick a capability to see what I bring.
The centre
Model the business first, keep one source of truth, and let agents do real work while people make the calls. Every capability around it serves that idea.
Five principles I build by
05How I build
I write flows and cut them into tickets. Claude Code agents build each ticket in its own worktree, two reviewers read every change, and I decide what lands. It is the loop I build with, so it is the loop I teach in CDA. Follow the boat.
Every feature starts as a flow: who wants what, step by step, written down before any code exists.
Spec firstAn automatic check rejects a flow that breaks the rules, long before it reaches a developer or an agent.
Automated gateOnly what changed becomes work. Nothing more, and nothing invented along the way.
Small ticketsAgents build each ticket in isolation, side by side, until every check is green.
Parallel agentsEvery change is read twice: does it do what was asked, and does it follow the house rules. I decide what gets fixed.
Two reviewsSmall changes, tested on every layer, in a history anyone can read back.
Clean history05.1The crew
Every change passes four stations, and nobody grades their own work.
01
Research
Facts first. One reads the code, one the documentation, one the web.
02
Build
The change itself, each ticket in its own isolated copy of the project.
03
Review
Two readers who never built it: one checks the ask, one the house rules.
04
Decide
Me. I read both reviews and decide what lands, and what goes back.
One production repository in its first ten weeks, counted from git and GitHub.
06Craft
I teach the image side of AI the way I practise it: reference first, then a structured prompt, then a hard audit of every frame. The same method runs through CCA. These are studies I made to push the method.
Self-initiated studies, not commissions. Each one starts from real references and ends as stills and short motion clips. Hover a frame with motion to play it.
06.1From a picture to an object
A clean reference image becomes consistent views, a generated mesh, and a cleaned, correctly scaled asset in Blender, scripted from the terminal. These are the real models, live in your browser. Pick one and drag to turn it.
Jade monument
07Slowly into focus
The photography behind Like A Human is mine: one theme, the world at the moment just before it makes sense. Light in transition, a form half revealed, mist about to lift. Two looks, one after Todd Hido and one after Saul Leiter. Every frame made in Midjourney, every subject named, every reject kept out.
From the Like A Human photography library
Accretion ring
01 / 12
07.1Then they move
Each loop starts on its own still and ends where it began, so it can run forever on a hero, a reel or a workshop screen.
08The lab
The lab is where I push the build loop furthest: a whole concept from one brief. Most of it is still half carved. That is the point: every run teaches the loop something new.
RTS concept
Scene study
Environment study
Agent research
Interface concept08.1Motion
Short vertical pieces from my motion framework, built in code with Remotion and HyperFrames, so the timing is exact to the frame.
08.23D experiments
DJI Avata, live
A drone rebuilt from product photos: a generated mesh, a decal pass for the markings, served as live 3D in the browser. Drag to turn it.
Obsidian Heart
A multi-shot cut that Blender renders and edits from a written shot list, without anyone opening the Blender interface.
A nature kit
Trees, roots, ferns and rocks for a real-time scene, each one cleaned, scaled and lit the same way, then rendered by script.
Materials
Physically based surfaces for the same scene: moss, lichen, stone, weathered wood. Each one checked on a sphere before it touches the world.
08.3Pages in a day
AI imagery, a hero film, sound and scroll motion, shipped as static pages. Concept pages, each from a single brief.
09In the room
Most of my weeks end in front of a room: an in-company team, an open workshop, or a stage at an official Anthropic Claude Code event. Everyone builds with their own tools on their own work. I build next to them.
Shortened from public reviews of Like A Human workshops. Pick a note up and move it. Read them all on likeahuman.ai
I now feel comfortable building in a terminal with Claude Code, after just one day. The sky is the limit.
Amazing to see how seamlessly everyone could jump in, regardless of their starting skill level.
I was able to get a deployment-ready version of my app with no background in coding.
Energetic trainers who can convey complex material in an accessible way. No long theory, just doing.
They really took the time to understand our business and objectives.
Rich in content, a good introduction to Claude Code and multi-agent systems in order to build.
10Contact
A workshop for your engineers, your business teams or your creatives. A talk at your event. Or a hard problem with agents that you want a second pair of eyes on.
academy@likeahuman.ai
