A few favourite posts from X and Instagram, where I share what I'm learning about AI, software and building products in public.
When an AI agent can generate hundreds of lines in minutes, reading every line becomes a poor use of senior engineering time. Define the expected behavior first, automate the checks you can prove, and save human review for architecture, trade-offs and failure modes. The goal isn't less accountability. It's better accountability.
Define what the code should do before it is written: clear acceptance criteria, recurring review comments turned into automated rules, behavior verified with evidence, and human reviewers kept on the decisions that matter.
Global fintech startups raised $10.6B in Q3 2026, down 21.4% year on year, across the fewest rounds since 2018. Investors now want traction, growth and a path to profitability, and AI that delivers measurable results, not just an AI label. (Source: Crunchbase News.)
AI looks cheap because VCs are footing the bill. After the model race comes the energy race. Whoever controls power controls compute.
Go deep, not wide. Learn relentlessly. Work hard on things you enjoy. Build relationships with people who challenge you. Create things even when the outcome is uncertain. Learn from people who've done it. Don't wait for the perfect time to start.
Yeah, I know, I'm late. But "better late than never" beats "never". What I'm building with it: the one thing I'm actually good at, turning other people's problems into products. Simple. Not easy.