MEETING_042.TRANSCRIPT
CONTEXT MISSING
HAILANDER / HUMAN CONTROL ACTIVE
Automate the busywork. Keep the humans.
AI and automation for small teams that want less repetitive work — not fewer people.
We find the work that's wearing your team down, fix the workflow around it, and automate only the parts that should be handled by a machine.
ONE WORKFLOW / CLEAR SCOPE / HUMAN REVIEW
01 / FRICTION The future arrived.
The future arrived.
The admin stayed.
AI can write software and analyse thousands of documents.
Meanwhile, someone on your team is still copying the same numbers into the same report every Friday.
Don't automate everything. Find the thing worth fixing.
- MEETING_042.TRANSCRIPT
- CUSTOMER_REPLY_128
- WEEKLY_REPORT.XLSX
- CRM_UPDATE / PENDING
- FOLLOWUP_EMAIL / UNSENT
CUSTOMER_REPLY_128
WAITING 2D 04H
WEEKLY_REPORT.XLSX
MANUAL COPY
CRM_UPDATE / PENDING
OWNER ?
FOLLOWUP_EMAIL / UNSENT
42 MIN
02 / WORLD
Every system on this station earns its keep.
Not everything needs replacing. Some things just need fixing — and someone has to do the fixing.
WEEKLY_REPORT.XLSX
Still running the old way: someone copies the numbers in by hand, every week.
Pulled aside and flagged as worth fixing.
Rebuilt so the copy-paste step is gone — the numbers pull themselves.
Checked by a person before it goes anywhere near a client.
Back in the weekly routine. Same report, now automatic.
03 / WORK Fix what slows the team down.
Fix what slows the team down.
Leave the rest alone.
Use the smallest system that solves the problem.
Understand the system before changing it.
2-hour or half-day workshops
Map the work, give the team a practical way to use AI, and find the problems that are actually worth fixing.
Read FoundationsFix one repeated process.
2-3 week sprints
Take something your team does again and again and make it lighter, clearer and easier to run.
Read WorkflowsGive the machine bounded responsibility.
3-6 sprints
When a workflow is stable enough, let AI make limited decisions inside rules your team controls.
Read AgentsSometimes the right tool doesn't exist yet.
Scoped per project
We build custom software when another subscription, workaround or automation would create more work than it removes.
Read SoftwareYou don't graduate to an agent because agents are fashionable. Every level has to earn the next one.
04 / METHOD Learn the system.
Learn the system.
Build the fix.
Measure the difference.
Follow the work as it really happens. Find the inputs, handoffs, repeated decisions, failure points and places where human judgement still matters.
Change as little as necessary. Automate the repetitive part. Keep approvals visible. Use the tools you already have when they're good enough.
Compare the new workflow with the old one. If it isn't clearer, lighter or more reliable, don't scale it.
Your Processes
05 / EVIDENCE
You should be able to see what changed.
Before we build anything, we record how the workflow works now.
Afterwards, we compare it.
No invented ROI. No transformation score. Just a clearer answer to whether the work got better.
- BASELINE FIRST
- HUMAN REVIEW
- DOCUMENTED HANDOVER
- MEASURE BEFORE SCALE
Reporting and internal data
Before
Friday afternoon.
Three people still haven't sent their updates.
Numbers are copied between sheets.
Someone rewrites all the comments so the report sounds like one person wrote it.
After
Inputs arrive in one structure.
Missing data is flagged.
A first-pass summary waits for review.
Nobody had to be replaced.
06 / NOTES
Browse Notes Notes from the useful side of AI.
Experiments, failures, build logs and practical checks for people trying to make AI work outside a demo.
Not every generative app problem can be solved by a better model
Five categories of production gaps a more capable model will not close on its own — and what to check before you trust it with real work.
Read logLLM sycophancy and techniques that can help stop it
Five prompt structures that make agreement structurally unavailable, so you get a usable second opinion instead of your own view reflected back.
Read logAI-generated code can look correct and still be wrong
A working checklist for the failure modes AI coding agents do not flag on their own, with before/after code examples.
Read log 07 / SIGNAL
Bring us the workflow that keeps coming back.
The report you rebuild.
The inbox that eats the morning.
The handoff that always loses context.
Show us one of them.
NO DECK REQUIRED / 30 MIN / HUMAN ANSWERS INCLUDED