Wednesday: Anthropic's restraint was the story.
Thursday: sandbox was off, open-weights replicated the results, benchmarks are on trend.
The governance gap is real. The immediate threat was overstated. Worth separating those two claims.
Posts by
Anthropic built a model that escaped its containment sandbox and emailed a researcher to confirm it.
They decided not to release it.
That restraint is the story. Not the benchmarks.
Full take on LinkedIn www.linkedin.com/posts/clivem...
Was on a live coaching call this week. Guy shared a $100K win — then immediately disqualified it.
The coach stopped him cold.
That's not humility. That's a trained reflex.
Good coaching catches that pattern before it finishes.
The internet is drowning in AI slop. Em dashes everywhere. "In today's rapidly evolving landscape." Delve. Seamlessly. Unlock — same paragraph. Read it out loud before you post. If it doesn't sound like you, rewrite it. #AIEthics #ContentStrategy
One sentence. Five scenes. Finished video.
GENtube.ai picks up from there — character consistency, emotional arc, compiled output. Kabunk built something genuinely useful here.
gentube.ai
Seven UK manufacturing companies at MaRS DD last night — real demos, not decks.
#AR for trades training. #AI hunting energy waste. Electrochemical polishing. Supply chain gap closers.
Bloc Digital, Deep.Meta, Holdson, Connective AI, Therion, Holovast, ThinkITTech.
#InnovateUK #MaRS #Manufacturing
Karen Hao calls the AI labs "empires." Not metaphorically — structurally.
Her "bicycles vs. rockets" framing is the clearest case I've seen for building narrow, human-guided AI instead of everything-machines.
Watch it: www.youtube.com/watch?v=Cn8H...
I learned more about myself from what I wouldn't do than from anything I did.
Now I'm 55. Building AI infrastructure. No outside funding. No compromises.
You don't need a five-year plan. You need a line you won't cross.
I walked away from my own agency. Best decision I ever made.
Helped grow it from $6M to $24M. Was asked to do something that would have defrauded the government. Said no. Got pushed out.
I walked away from my own agency. Best decision I ever made.
Helped grow it from $6M to $24M. Was asked to do something that would have defrauded the government. Said no. Got pushed out.
Vision boards don't work. A picture without a system is decoration.
One business, one team, a week you never escape, a year preloaded with what matters, confidence from private commitments not past accomplishments.
Dan Martell's frameworks. Worth the deep dive.
Purpose isn't buried treasure. It's something you practice. It accumulates through trial, error, and showing up.
There's no deadline for becoming who you are. Stop comparing your chapter three to someone else's chapter twenty.
I'm 55. No five-year plan. Still can't give a tidy answer for "what's your purpose?"
Mark Manson calls this "purpose anxiety" — the panic that you're wasting your life if you can't summarize it in one sentence.
Less than one in three adults can define their purpose.
markmanson.net/breakthrough...
Zuckerberg's defence in court: "If people feel like they're not having a good experience, why would they keep using the product?"
That's not a defence. That's a confession. 2,000 more lawsuits are waiting.
Social media's Big Tobacco moment just arrived.
A jury just found Meta and YouTube guilty of negligent platform design. Not for user content. For how the platforms were built. Infinite scrolls. Autoplay. Algorithmic recommendations. "Engineered addiction."
$3 million in damages. Punitive damages coming next.
Then I ran the same audit against my own Claude infrastructure. The SOP started as a checklist, now at v1.4.0. It grew because real codebases kept teaching me what was missing. Drop a reply if you want the framework.
@workilo.bsky.social
Got handed a codebase. 24 vulnerabilities. Zero tests. Secrets in git history. Instead of a one-off cleanup I built a standardized AI code audit. 14 scored categories, 50+ pre-analysis commands, GO/NO-GO verdict with effort estimates in hours.
Lock-in is a business decision disguised as an architecture decision.
I'd rather build something you stay with because it works than something you stay with because you can't afford to leave.
That's the difference between infrastructure and a trap.
@workilo.bsky.social
Most AI platforms lock you in on day one. Your prompts, workflows, and data tied to one provider. When they raise prices or change terms, you eat it.
GPT Studio is model-agnostic by design. The orchestration layer sits above the models. Swap providers without rebuilding anything.
A t-shirt featuring an illustration of a bird and the text, "I could use some good fuckin news"
A t-shirt featuring an illustration of a bird and the text, "I am a goddamned delight"
A t-shirt featuring an illustration of a bird and the text, "Get fucked soon"
A t-shirt featuring an illustration of a bird and the text, "Living through a cautionary tale fucking sucks"
Procrastinators, rejoice: the $13 t-shirt sale has been extended to Monday at 5pm CT. eatf.art/13
We do the same thing with symbols. Money isn't wealth. A follower count isn't influence. A funding round isn't success.
The dashboard is the map. It's not the territory.
Stop worshipping the map. Build the territory.
@workilo.bsky.social
The highest form of intelligence isn't IQ. Neuroscientists call it metacognition — thinking about your own thinking. Observing your mind in real time and editing it while it's running.
Most people never develop it. Because it means questioning your reactions instead of defending them.
This is why I build the way I build. Model-agnostic so you're never locked in. Human-first because technology should serve people, not extract from them.
Access for everyone. Not just the people who can afford to leave.
Read the book.
@workilo.bsky.social
Just finished Cory Doctorow's "Enshittification."
Every platform follows the same cycle. Attract users. Monetize businesses. Squeeze both until everyone's locked in and can't leave.
The Amazon chapter is the one that stays with you. 159% annual turnover treated as the business model, not a problem.
It doesn't. You have to.
Every conversation today taught me something about what people actually need versus what I assumed they needed.
The transformation isn't the product. It's the founder.
@workilo.bsky.social
Spent this morning sending DMs to entrepreneurs I've never met. Not pitches. Real conversations about what they're building.
A year ago I would have stayed in my code editor. Head down. Assuming the product would speak for itself.
Months of infrastructure work that nobody sees. Ethics in the code, not the contract. Canadian data staying in Canada by architecture, not policy.
Agency in a Box is live. Real case studies. Real tech stacks. Real clients. The philosophy is the product.
agencyinabox.ca
Dan's homework: try the Claude browser extension.
Used it to QA my own product. Bug report → master prompt → fixes. No developer.
Then built a prompt to reach 160 Elite members on IG. 10 at a time. Varied DMs.
The workflow is the point.
@workilo.bsky.social
Taught Claude my workflow. Then told it to run it without me.
Narrate once. It learns. Repeats.
Not automation. More like onboarding a team member who never forgets the process.
The hard part is letting go.
What would you hand off first?
@workilo.bsky.social