The blog.
Deep dives, comparisons, and roundups on the AI tools worth paying for. Only what we'd actually buy — no SEO sludge, no influencer demos.
Best AI search engine: one question, three engines
The best AI search engines, tested with one identical question. All three agreed on the facts. One cited EUR-Lex, another a consultancy blog and YouTube.
Claude vs ChatGPT vs Gemini: they don't all cost $20
Claude vs ChatGPT vs Gemini, after reviewing all three separately. Two cost $20, one is a storage plan — and the free tiers differ far more than the paid.
Best AI voice generator: what an hour of audio really costs
ElevenLabs wins on voice and costs $10.91 an hour. We normalised six AI voice generators plus open-source to one unit: the cost of an hour you can publish.
Descript pricing: two currencies, and one tier that costs more
Descript pricing runs $0 to $65 a month across five plans. Read live today — including the tier that costs more per media hour than the one below it.
Copilot vs Codex: you may already be paying for both
Copilot vs Codex, both measured on one identical brief. People ask whether Copilot has Codex — it does, on the $10 tier, which changes the maths entirely.
GitHub Copilot pricing in 2026: what one AI credit actually buys
GitHub Copilot pricing runs $0 to $100 a month, but since June 2026 you buy credits, not seats. Here is what a credit buys and how fast it goes.
Claude Code vs Copilot: $10 now buys you both
Claude Code vs Copilot, both measured on one identical brief. Copilot Pro at $10 now bundles Claude Code access, which changes the question entirely.
Claude Code tips: what eight months of daily use taught me
Claude Code tips from running a production pipeline on it daily for eight months, plus a controlled build test. The habits that changed my output.
Lovable vs Replit: same prompt, same six requirements, two bills
Lovable vs Replit on one identical brief. Both hit 6 of 6 — Lovable in 109 seconds, Replit in four minutes and 31% of a day's credits. The cost model decides.
Antigravity vs Claude Code: I gave both one brief, neither built it
Antigravity vs Claude Code, measured on one identical brief. Both refused to build unattended — the opposite of what page one says. Costs, code and verdict.
Google Antigravity review: it wrote its own tests
Antigravity built my six-requirement app on the free tier and met all six. Then it wrote and ran its own unit tests, which no other builder I have tested does.
Lovable pricing: what a credit actually costs
Lovable pricing decoded: one credit costs 25 cents, a measured real build cost thirty cents, and Business charges double per credit for the same 100 a month.