Writing about AI, technology and commercial growth: what worked, what failed, and what helped me move from pilot to production.
Nathan Petralia
I have spent two decades leading digital and commercial programs across APAC. Today I build AI products and leverage AI across consulting, practice building, go-to-market, commercials, delivery governance, and operational leadership.
About me →Recent Posts
Do not build Phase 2 orchestration until Phase 0 data says so. Layer 4 feedback — CSV, footer Agents line, eval gate — plus weekly OpenRouter checks beat benchmark leaderboard anxiety.
Continue reading →
External memory is four tiers in practice — short-term, operational, evergreen, and a feedback loop hardened into rules and footers. The harness gates when each tier loads so you keep control without token bloat.
Continue reading →
Terminal-Bench harnesses look like separate products. On a production Shopify app I already had subagents, CI gates, and session rules. You keep model and mode control — the harness supports routing, tests, and memory gates, not autopilot.
Continue reading →
Anthropic's Fable 5 leads CursorBench 3.1 at 72.9%, but at $18 per task and 76 steps. I read the table for score per dollar, tokens, and steps, and where open models land.
Continue reading →
I ran Headroom, built a 300-line proxy, wired a Cloudflare tunnel, and added RTK. On my Cursor + OpenRouter workload the dollars did not move. Here is what is worth doing instead.
Continue reading →
Copilot had the same footer spec but dropped it on long chats. Cursor keeps it with alwaysApply rules, optional hooks, and a v3.1 mode-based Response Footer Contract.
Continue reading →
Most enterprise AI programs struggle before the model fails. Data, governance owners, and change runway get compressed. A pre-flight diagnostic before the next funding gate.
Continue reading →
Five management habits that transfer directly to directing AI agents: show examples, write context down, guide in steps, define outcomes, and close the loop with review.
Continue reading →
A thousand-line profile block in one extension fights merchant menu IA. Split full-page extensions by job and align with how customers navigate account tasks.
Continue reading →