HowTo AI Content · Björn Sennbrink · Sweden
I help Nordic teams get there with clear words, repeatable instructions and honest tests. I use AI every day, and I built ALTRUIXM with it.
In practice
Field notes
The shortcuta magic wand
In practice
A set of repeatable instructions, often with scripts and examples, that an assistant loads when a task calls for it. It is a process document the model can follow.
Why it mattersInstructions can be written, versioned, reviewed and tested. Magic cannot.
The shortcutan app
In practice
An open protocol: a standard way for AI assistants to connect to tools and data. A server speaks the protocol, and many different apps can use it.
Why it mattersA protocol draws the trust boundary: what the model can read, what it can change, and who approved it.
The shortcutwe’ll see when it’s live
In practice
A repeatable test of AI output against examples you define. It tells you whether a change made things better or worse.
Why it mattersWithout evals, every prompt change is a guess dressed up as progress.
The shortcuta digital employee
In practice
A model working in a loop with tools, acting on instructions within limits that someone set and someone owns.
Why it mattersEmployees carry responsibility. An agent carries the responsibility of whoever configured it.
The shortcuta growth mindset
In practice
Stakeholders saying yes to a vision of unlimited growth while overlooking the real organisational realities.
Why it mattersAI-fy human-centred processes without a plan, and sooner or later the workforce is put up against the machines. Plan the people side first.
The shortcuta bottleneck
In practice
The moment a person decides. AI prepares the decision: the options, the evidence, the draft. A person makes it.
Why it mattersIt is where accountability lives, and where quality is protected at scale.
The irony
We are handing over process development, the actual work and the decisions at a pace that would make an F1 race car envious.
350km/h
Speed is not the problem. Losing the steering wheel is.
01
One person who understands what the AI does there, and who can say stop.
02
Evals and a small pilot before AI takes over a whole process, not after.
03
Agree in advance which signals make you slow down: quality, risk, people.
How I work
Role 01
It drafts, researches, structures and codes, tirelessly. I give it clear instructions and review everything it hands back.
What I ownThe brief and the review.
Role 02
At its best it questions my plan, points out what is vague and suggests another path. When the suggestion is better, I take it.
What I ownThe decision.
In both roles, a person decides.
Proof · ALTRUIXM
ALTRUIXM is a desktop app that analyses existing content, finds the gaps and writes from them. It works, and I use it every day.
I built it by describing the result I wanted and staying in charge of every decision. When the AI suggested a better path, I took it. Then I tested: prototypes, evals and user tests, round after round.
The loop: Describe the result, then AI proposes, then I decide, then Build, then Run evals, then Test with users, and around again.
For leaders
Ask
If nobody can write the instructions down, it is not ready. Vague scope is the most common risk, and the cheapest one to fix.
Ask
Which systems the model can read, what it can change, and what leaves the company. Protocols like MCP make this explicit. Use that.
Ask
Agree on evals before launch: examples of good output and a test you can rerun after every change.
Questions like these are how a pilot becomes a product.
Latest writing
If AI writes the whole content chain over time, output converges on earlier AI output. What I changed in ALTRUIXM, and why.
Read on LinkedIn →