Practical writing on using AI well

I'm Nathan Petralia — operator and writer on enterprise AI, file memory, and getting agents to production.

What worked, what failed, and what helped me move from pilot to production.

New posts ship on the blog; the weekly digest catches what you missed — one email, practical only.

Nathan Petralia at HKU

Nathan Petralia

I have spent two decades leading digital and commercial programs across APAC. Today I build AI products and use agents daily across consulting, practice building, go-to-market, delivery governance, and operational leadership.

About me →

Start here

New to the site? Pick one path — each opens a guide or series hub.

Browse all writing →

Reading paths

Multi-part guides — start at the hub, then follow the series in order.

Latest

New this weekCinematic 16:9 low-angle of a glass library with amber-lit shelves and light beams through dust.
Strategic

GEO Is Site Legibility for Agents

Best forMarketing and content leaders who need structural work that helps AI systems cite your site accurately

GEO is passage-level clarity so retrieval systems can quote you accurately. Here is how site legibility complements classic SEO and AIO.

Continue reading →
New this weekCrawl map on a wall with three glowing paths to abstract sitemap, robots, and llms icons, streetlight and moonlight, no readable text.
Hybrid

IndexNow, Sitemaps, and Why Agentic Browsing Cared About llms.txt

Best forAnyone publishing a personal or company site who wants traditional SEO discovery and GEO llms.txt to stay in sync without treating either file as a gimmick

Search crawlers read sitemaps. AI crawlers read llms.txt. Lighthouse Agentic Browsing now audits both. I wired IndexNow, sitemap regen, and llms.txt into one publish chain on petralian.com, including a date-parsing fix the audit exposed.

Continue reading →
Cinematic 16:9 of a tuning fork vibrating beside a typed card on a walnut desk.
Hybrid

Brand Voice in YAML: One Source for Humans and Agents

Best forMarketing ops leads who already govern voice in documents and need the same rules readable by agents without duplicating prose guides

A PDF voice guide agents never open fails at scale. YAML fields for tone, banned words, and claim limits give humans and models the same machine-readable source.

Continue reading →
Conference room before standup with a single clipboard on the table and chairs angled inward.
Hybrid

Session Bridge as the Standup File Agents Read First

Best forDelivery leads and program managers who want agents and humans aligned on sprint intent before standup, not after transcript archaeology

Standup worked for humans in the room. It never fed the next session. A short Bridge file after standup gives agents and teammates the same priority, blockers, and owners.

Continue reading →
Open research journal with abstract diagrams beside an organized workshop bench in morning light.
Hybrid

When Poe Is the Research Layer and Cursor Is the Build Layer

Best forAnyone who researches in Poe or other chat tools and wants file-grounded builds in Cursor without context bleed

I run open-ended research in Poe, promote decisions into files, then use Cursor to implement in the repo and vault. Splitting layers beats one chat for everything.

Continue reading →
Steel filing vault with one drawer open showing glowing data cards instead of software logos.
Hybrid

The CDP Your Agents Need Is a Folder Contract

Best forMarketing and program leads running agent pilots who need governed customer context without buying another CDP seat

Composable martech plus a folder contract: governed customer-context paths for agents—when CDP, lake, or hybrid stacks fit, and best practice for commerce and marketing.

Continue reading →
Browse all writing →