A deep dive into how I work — the new way of thinking, the models, the automation, and why imagination is now the only constraint that's left.
We were trained to think in one direction — task in, output out. AI breaks that. The new edge isn't knowing more tools. It's learning to think with them. That's a different cognitive shift entirely.
The tools are here. Anyone can access them. What separates the outputs is the quality of the idea going in. Creative thinking — not prompt engineering — is the new competitive moat.
A sketch on paper becomes a film. A voice note becomes a published essay. A physical installation becomes a data sculpture. The boundary between making something real and making something digital has collapsed.
Scheduling, formatting, summarising, filing, drafting the same email for the tenth time — this is now automatable. The question isn't whether to automate it. It's what you'll do with the time that was wasted on it.
Not a rough cut. Not a demo. A real short film, a series, a visual album. The production barrier is gone. The only remaining barrier is vision — knowing what you want to say and having the nerve to say it.
The AI does the execution. You do the taste, the judgment, the narrative. This is actually harder than writing the code yourself — it requires clarity about what you want, which most people don't have and have never needed to develop.
The best ideas come out in conversation. Talk to the model the way you'd talk to the sharpest person you know. Ask the uncomfortable follow-up question. Push back. The quality of your thinking shapes the quality of the output.
The AI doesn't think for you. It thinks with you. Curiosity is the engine. Questions are the fuel. The answers build themselves.
Most people skim the surface of every tool. The real edge comes from going deep on a few. Understanding why a model behaves the way it does gives you leverage everyone else misses.
No single model has the full picture. Use Gemini for one thing, Claude for another, DeepSeek for a third. Cross-reference. The tension between answers is where insight lives.
All the data, all the copy, all the reports — they can already be built. The only thing that separates what gets built from what doesn't is whether someone imagined it clearly enough to ask.
All the data, the copy, the reports, the finance, the automation — it can all be built. The only question is whether you can picture it clearly enough to ask for it.
Vague questions get vague answers. The more precisely you describe what you want, the more precisely AI can deliver it. Clarity is the skill.
If one model doesn't get it, another will. Each has a different relationship with language, logic, and knowledge. Expertise is knowing which to use when.
Surface-level prompts get surface-level results. The people winning right now are the ones who go three levels deeper than everyone else in the same conversation.
The best solutions fix your own problems. If you deal with it, other people do too. That's the only product thesis you need to start.
Focus beats comprehensiveness. A small tool that works is worth more than a grand vision that never ships. Build the tracker, the reader, the dashboard — something you'd actually open tomorrow.
Paralysis comes from too many options. Pick the one thing that removes the most friction from your actual life. Not the most impressive thing. The most useful one.
Every feature you add before the first version ships is a reason it doesn't ship. Small and done beats ambitious and perpetual. Ship it. Then improve it.
The best products are scratching your own itch. You already know the edge cases. You already feel the friction. You're the most honest user you'll ever have.
Not a hypothetical. Not a market opportunity. Something you'd use today if it existed.
If you can't describe the first version in two sentences, it's too big. Reduce it until you can.
If yes to all three — build it. You're probably not the only one with this problem.
Building is better than knowing. · Information is cheap now. · Building is the best way to learn.
Real things people are building with agents right now. Not demos. Not prototypes.
This is the landscape. Every category below has people shipping real products, alone or in tiny teams, that would have required 10+ people two years ago.
Short films, visual albums, brand films, social content series — all produced solo using AI video, voice, image generation, and editing agents. Physical direction + digital execution.
Lead research, personalised outreach, follow-up sequences, CRM updates, proposal generation — all automated. Founders closing enterprise deals with zero sales headcount.
Daily briefs, weekly synthesis, rolling signal feeds, radar pages — all generated, curated, and published automatically. One person operating what used to be a 5-person editorial team.
Live portfolio monitoring, signal scanning, automated trade execution, and daily finance briefs. From paper trading systems to live algo strategies — running 24/7 without a desk.
Habit trackers, gym logs, personal dashboards, sleep optimisers, mood journals, focus timers — built exactly how you want them, with exactly the features you need. No compromises.
Bilingual curricula, AI tutors, workshop decks, community programmes — built and iterated without a content team. One person running what used to be an education department.
Knowing which model to use for what is the real skill. The field is no longer locked to a few providers. Open source has caught up.
Current routing logic — updated March 2026. Dynamic task classification selects the right model automatically.
Not a chat interface. Not a browser tab. A persistent AI agent that knows me, learns from me, runs while I sleep, and talks to me through my phone.
Every message is a command. Every reply is intelligence. No login, no context-switching — a conversation that never ends and always remembers.
Daily logs, long-term memory, instincts learned from sessions. Knows my projects, repos, preferences, rules. Updates itself. I never repeat context.
GitHub, Notion, Alpaca, web search, X, email. Ask in plain language — it finds the right tool, runs it, reports back.
Sits quietly in team Telegram groups. Only speaks when mentioned. Reads everything to build context. The team can use it too.
Article writing, market research, investor outreach, GitHub automation, image generation. Each one a structured process loaded on demand.
Every day, a pipeline runs automatically. Research gets done, briefs get written, posts get published, signals get scanned, patterns get synthesised. I wake up with intelligence, not tasks.
Reads 10+ sources, synthesises top stories, publishes intelligence brief to site, feeds radar.
Live portfolio from Alpaca, X signals from 30 followed accounts, finance brief delivered to Telegram.
Reads a full week of signals across AI, geopolitics, cyber, capital, culture. Surfaces 3 patterns with 1st / 2nd / 3rd order consequences.
Scans X for viral AI / tech posts. Surfaces what's moving in real time. Delivered to group Intelligence channel.
Curates best long-form from 10 RSS feeds — philosophy, capital, geopolitics, AI, culture. Scores by signal quality, not recency.
Synthesises the week of radar items. Claude writes the weekly digest — headline, pattern, watch-next. Publishes to site.
A website. A dashboard. An automation. A signal pipeline. Start with one thing. Be curious. Ask the right question. The rest builds itself.