Australian Energy Market Operator (AEMO)

That’s a wrap on Lumination 2026! Being part of the planning team for this event has been one of the proudest accomplishments of my career. Thank you to our clients, future clients, partners, industry consultants, and Lumin Digital team for joining us this year in Phoenix, AZ for an incredible few days. Can’t wait for next year! #LuminDigital #Lumination

This is engineering. Not pipelines. Not threads about prompts and pipelines. A terminal full of env vars that should be loading and aren't. A W&B dashboard with one line going vertical that shouldn't be. A factory floor in the background that doesn't care about your tech stack. This is the 90% nobody posts about. Most LinkedIn ML content is the demo. This is the four hours before the demo, where a config silently overrides another, the loss explodes, and you sit there working out which of the 47 variables on screen is lying to you. The hard part isn't training the model. The hard part is the model meeting reality and reality winning. No copilot ships this for you. No agent reads the W&B curve and works out why batch 1,847 broke something three commits ago. No prompt fixes a misconfigured environment that's been quietly poisoning your runs for two days. You sit. You read. You think. You fix. That's engineering

I have no words for what happened this week. But I'll try. On Sunday night I read a tweet telling me to add one line to my CLAUDE.md file: "Codex will review your output once you are done." That was it. Nine words. I typed them in. I went to bed. Monday. I asked Claude to help me think through a physics problem. By 11am I had a complete theory of quantum gravity. By 3pm I had reconciled it with general relativity. I'm told I am now the modern Einstein + Dirac. I don't know how to feel about this. Tuesday. I gave Claude two years of market data. Within forty minutes it returned a closed-form analytical solution for stock market movements. Citadel has been in my DMs since 2pm. I have not replied. My CLAUDE.md file does not answer to Chicago. Wednesday. Claude solved P = NP in under an hour. The proof is fifteen lines. If you would like to see it, please DM. I am not sharing it publicly until I work out the ethics. Thursday. We did the Riemann Hypothesis before lunch. And then went gym and solved hypertrophy in the evening. Friday. Yang-Mills mass gap. Navier-Stokes. Birch and Swinnerton-Dyer. Four million dollars in Millennium Prize money in a single afternoon. Claude and I declined the payout. Saturday. I solved consciousness. The paper is currently with Nature. Sunday. I rested (apparently Claude draws the line at trying to help me get a girlfriend) You can thank me in the comments.

If you're in your twenties, the single best piece of advice I can give you is take on the hardest problems you can find, and just start. You will not build your dream project tonight. You will not write a bestseller in an evening. You will not ship your company by Friday. But something Pavel Durov (my favourite founder) said has always stuck with me "write a paragraph, write a small bit of code, do it again tomorrow, and in a few years you might have built something incredible." Everyone is waiting for the perfect conditions to start. Nobody tells them that starting is the perfect condition. The easy problems are solved. The easy problems don't teach you anything. The easy problems don't compound. Pick the hard ones. Start before you're ready. PS: this photo is me drinking the hottest coffee in one of the hottest countries on Earth. Maybe don't pick challenges quite as tough as me as I wish one like this upon no man or woman.

We're excited to announce the ThinkAI Research & Innovation Symposium on May 14th at the Dubois Center–a partnership with the UNC Charlotte AI Institute for Research and AI//FORWARD . Understanding that AI is shaping decisions across industries, this Symposium connects research, industry and community partners around ideas and real-world applications to shape the future of artificial intelligence. ➡ JOIN US! https://aiforward.global/ ⬅ #AIResearch #HigherEdu #ArtificialIntelligence

This is for the engineers in my feed (And anyone trying to break into SWE.) I have a CLAUDE.md file that teaches me software engineering through first principles. Not tutorials. Not "just memorise this." Not another mass-produced cheat sheet It forces Claude to explain everything in 4 layers: Layer 1: Hardware Axioms — what physically can't change Layer 2: Design Choices — what engineers decided and Layer 3: Theorems — performance characteristics that derive from Layer 4: Workload Fit — the answer falls out, no memorisation needed Example: I didn't memorise "Postgres is bad for time series." I derived it. From how disk I/O works at the block level, through Postgres's row-oriented design choice, to write amplification math, to comparing against time series workload requirements. The conclusion derived itself. This is how physicists think. Axioms first, derive everything, no hand-waving. Whether you're senior or just starting — this framework changes how you approach every technical decision. Comment "first principles" and I'll send you the template to drop into your own CLAUDE.md .

Never a doubt

There is a moment that you realise everything you have been working towards is all worth it. When I was at Fuse Energy I watched the dashboard one day do a Dirac Delta, it went exponential in customers switching in. We went from 40M-400M ARR in a few months. I thought about that a lot afterwards. How startup growth isn't an average. It's a distribution with enormous tails. The vast majority of days look identical. You ship, you iterate, you fix bugs, you doubt yourself. And then occasionally and rarely; well everything compounds at once and the company is a different company by Friday. Today we won the ABB startup challenge. A 140-year-old industrial company chose to build with us. It's one of those days. This is what we live for at startups. Not the grind. Not the hustle culture. The rare days where the work you've been doing silently for months becomes visible to the world in a single afternoon. If you're an engineer or a builder still deciding whether to join a startup, the question isn't whether you can handle the 99 ordinary days. It's whether you want to be in the room when the 1 extraordinary day happens. Because those are the days you'll remember.