Now

Building Slow moved to Substack

The newsletter now lives on Substack as Building Slow. The first three long reads went out this week: the four files that replaced my freelancers, why a mistake log beats hunting for the perfect prompt, and why I stopped selling WordPress sites.

The rule I'm holding myself to: only write about things I've actually built and broken, with the real numbers. No schedule, no filler.

No more new WordPress builds

Since 5 September, Bingo doesn't build new WordPress sites. New builds are plain files on Astro and Cloudflare, because more and more of the work in my agency is done by AI, and AI is good at files and bad at dashboards. Existing WordPress clients stay where they are, and we keep looking after them.

The same shift happened with people. Work I used to send to freelancers is now done by Claude, run from a handful of plain text files. And it hasn't been a smooth year: I've lost 11 of my 62 regular clients in 2026. Writing that down next to the wins.

Client work right now

• KKIMD (DBKK) — daily site updates while the event runs.
• A premium wildlife travel brand — getting ready to launch.
• A food business — fixing its tracking, then writing it up as a case study.
• A new website offer — testing ChatGPT ads for a month, starting 6 October. Too early to say if it works.

My own builds

HelloReply, the AI receptionist that answers WhatsApp, Facebook and Instagram messages for small businesses, has its first paying customer. A second business is trialling it. It also has a deadline: if it isn't paying its own way by the end of March 2027, I park it.

GoReview is restarting. I'm testing it on my own florist first, before anyone else sees it.

Happy Florist is putting its energy into returning customers instead of chasing new ones.

Tools I've built to make this possible

Four industries on a team of under five only works because the repetitive parts stopped being done by hand:

• BingoOps — the agency operations dashboard. Tasks, checklists, close-out tracker, work log, client hub.
• ZeroRank — tracks how a brand is described and cited across seven AI engines over time.
• A public pricing calculator — real project estimates in about a minute, so first calls aren't spent on budget.
• SabahGuide's event pipeline — scrapes local news and listings, LLM-extracts real events into one approved queue, and that queue feeds both the site and the newsletter. Turned a weekly writing job into a weekly reviewing job.
• Bagasi's page pipeline — generates and cross-links 1,800+ hub, guide and destination pages from structured data.
• An outreach system — self-submission link building and sequenced follow-up. Unglamorous, necessary.

A fair criticism of that list: it's a lot of machinery for a person whose main problem this year has been not shipping. Building tools is the most sophisticated form of procrastination I know. But the ones above genuinely bought back hours, and the newsletter you're being asked to subscribe to is largely about which ones did and which ones didn't.

The long plays: SabahGuide and Bagasi

SabahGuide is past 1,000 pages. Bagasi is past 1,800. The lesson from this year still stands: pages built is not readers earned.

Family

The kids walk to the padang on their own now, no parents hovering. They've made local friends and are speaking real Bahasa Malaysia with them — not the classroom kind. Watching that happen has been the best part of this year.

Mornings belong to them and to the gym, not to WhatsApp. That rule has held better than most.