Impact Lab · Track 1
Creative work under pressure
Creative work in LA is being repriced, restructured, and in places disappearing — and the people it is happening to have less information about it than the people doing it to them.
Someone wrote
At the July conversation the room wrote on cards. These are unsigned, and quoted exactly — spelling and all.
“CREATIVES/KNOWLEDGE WORKERS FEEL IT THE MOST.”
“Insecurity… are our ideas realey ours to own?”
“Expectations of Output are Changing. You should 2x or 3x yourself and broaden scope.”
“So much new knowledge + capability, how can we feel empowered not replaced?”
“WE all know we need to adapt, can we adapt in time?”
“Burnout”
Written on cards at a Claude Conversation. Unsigned, and quoted exactly.
What people said
The fears clustered. Work drying up without explanation. Rates falling while the work stayed the same. Contract language nobody could parse. Being asked to move faster with no say in what that meant. Not knowing whether to retrain, wait, or leave the industry.
The hopes were narrower and more specific: knowing what is actually happening, in time to act on it.
The opening
Somewhere between those two is a gap you can build into. Some directions worth pulling on.
Comprehension over detection. Contracts, rights, terms — most people signing them cannot read them, and that is more tractable in a day than any provenance arms race.
Time and price, not headcount. Creative people are paid in days. Job-loss framing gets headlines; day-rate math gets used.
What moved, specifically. “Is this job at risk” is unanswerable. “Which of these eleven tasks moved” is answerable and more useful.
The lateral view. Where else these skills are worth something — locally, with real numbers attached.
Collective visibility. One freelancer’s rate drop is an anecdote. Many is a pattern, and nobody is aggregating it.
Data to consider
Categories, not files. Go find what is good.
Federal occupation and task taxonomies. Metro-level employment and wage series, which run multi-year if you want a trend. State layoff and unemployment filings. AI usage and exposure research, including Anthropic’s own published economic data. Guild, union, and trade-association reporting. Regional production and permitting activity. Contract and rights language, which is mostly sitting in people’s inboxes rather than in any database.
Assume anything you find is a few months stale and measured for a different purpose than yours. Say so on the slide.
Don’t
Don’t build AI-detection or provenance. Funded arms race, and a day build will be confidently wrong.
Don’t build a chatbot for the emotional part. The distress is real and a bot is the wrong artifact. Route to people and to actual services.
Don’t imply displacement from a correlation. Employment series are not built for year-over-year comparison and a drop has many causes. A signal is not a measurement.
Pick one creative occupation. Just one.
Before touching data: list what someone in that job actually does in a week. Ten to fifteen concrete tasks, in their words, not a job description’s.
Then ask, per task: has this moved? Who says so? How would they know?
Build for the tasks where the answer is clearest. Show your uncertainty on the rest — that honesty is the product, not a caveat on it.
Bring this to a lab
Impact Labs are one day, in a room, on a real problem. The next one is on the calendar.