The AI era · career call
01 / 08
One thing I'm increasingly sure of

The most valuable work from now on
is testing + safety

AI will keep getting better at doing things. But who makes sure it's right, low-risk and won't fail?

What will be worth most
isn't production, it's acceptance and safety nets.
Chapter · 01
02 / 08
Chapter 1

Start with the big trend

As execution gets cheaper, judgement gets pricier.

More and more work goes to AI
03 / 08

Within 3 years, all of this will run itself

Content
Copy · proposals · scripts
Creating is one prompt away
Production
Images · pages · spreadsheets
Productivity tools get rewritten wholesale
Execution
Workflows · code · automation
Agents move from demo to real workflows
Analysis
Read data · summarise · advise
Decision support reaches all the way down
Why this is happening
04 / 08

The stronger AI gets, the more judging right from wrong is worth

1
Production gets cheaper, marginal cost near zero
2
Copying gets faster, and errors speed up too
3
AI gets it wrong with a straight face, and humans struggle to notice
4
So the most valuable skill becomes spotting problems
AI makes building it cheap,
but makes getting it right, stable and safe pricier.
An acceptance Skill you can run tonight
05 / 08

Don't write prompts,
write acceptance checklists

# skills/ai-acceptance/SKILL.md
name: ai-acceptance
description: "Runs AI output through a 4-gate review checklist."

gates:
  - functional:  "Does it actually do what the user asked?"
  - edge_cases: "Empty / long / non-ASCII / concurrent?"
  - safety:     "PII, secrets, destructive ops — all red-flagged?"
  - rollback:   "If this ships and breaks, can we undo in 60s?"
Relative change in role value
06 / 08

Getting cheaper, getting pricier

Pure execution -65% value Content production -40% value Data analysis Flat Testing / QA +85% value Safety / risk +110% value
Three things you can do tonight
07 / 08

Stop chasing tools,
train your judgement

Tonight
Write an
acceptance list
Even 5 items will do. Starting beats perfect
This week
Run a
red-team drill
Tell your own agent: try to make it fail
This month
Add a
rollback process
If you can undo in 60 seconds, you can let go
What's truly scarce isn't knowing how to use AI,
it's daring to sign off on AI's results.
Thanks for reading
08 / 08
Thanks · thanks

If you're thinking about this too, tell me in the comments:
what do you most want AI to do for you? What do you trust it with least?

@lewis RedNote · White editorial style html-ppt · full-deck