Writing
Notes on the work underneath.
Posts on building AI that's reliable enough to put in front of customers — alignment, auditability, and the engineering you don't see in the chat window.
-
7 November 2026
A written brief instead of a voicemail: a clinic phone line through the same assistant
What happens when a patient calls a clinic line that goes through the assistant: a recorded greeting, a written brief for staff, usually a WhatsApp follow-up, and no AI on the call.
-
4 November 2026
Your clinic has been offered "AI search optimisation". What happens after a patient's assistant finds you?
Being found by ChatGPT or Google's AI is the first step. Where AI search and its ads stand in Singapore, what we build for the moment after an assistant finds a clinic, and three plain questions to ask whoever you buy from.
-
4 November 2026
Your client wants their AI assistant on another channel. How much of it changes?
The part that answers questions should not change at all. What changes is everything at the edge: what comes in, what goes out, and who the person is. What to test first, what to leave, and the questions to settle before you quote a second channel.
-
16 October 2026
PSG is gone. What the new EDGE grant means if you want to use AI
On 30 September three Singapore grants became one. What EDGE covers, why digital work gets up to 50% and not the 70% in the headline, who applies, the rule that loses the most claims, and where AI fits.
-
1 October 2026
Your client wants an AI assistant grounded in their data. Who keeps the data right?
The phrase covers at least three different builds, and only one of them looks the answer up from a named record at the moment it answers. The lookup is the easy part; the catalogue behind it is never finished. What to settle before you quote, and the questions to put to a vendor or to your own build.
-
16 September 2026
I built the MCP server twice, and the second one went through the middleware
The first one pointed its tools straight at the booking platform's API, and I switched it off within weeks: a model reasoning from scratch needs a dozen calls for one ordinary question, which is unusable against a tight rate ceiling. Rebuilding it on the middleware took a fraction of the effort, because that layer had already been shaped by the same constraint.
-
16 September 2026
What should a clinic automate first, and what should it never automate?
Start with the enquiries your staff answer the same way every day, where the correct answer is written down somewhere you control. Keep clinical advice, consent boundaries and undecided policy away from it. The sorting exercise, the failures that actually happen in clinics, and the questions worth asking a vendor.
-
26 August 2026
Before you hand your WhatsApp number to a vendor
"Just give us your number, we'll handle the Meta side" is the most consequential decision in the whole engagement, made in about four seconds. What a WhatsApp Business Account actually owns, where the patient conversations really live, what an exit costs — and the five questions to ask on the onboarding call.
-
24 August 2026
What has to be true before an assistant reschedules for you
A reschedule is a write to a live operation. Why a reschedule and a cancellation are different risks, why the scoped handover link is the design rather than a cop-out, one authority on what may move, and why stale intent has to expire on its own.
-
22 August 2026
When one appointment is actually several procedures
When one thing a customer books is several things the operation runs, the booking has to be the unit that stays consistent, partial success has to be treated as the normal case, and the seams underneath have to be genuinely closed rather than hidden in the wording.
-
22 August 2026
What changes when a bespoke integration becomes a reusable core
Most of what looked like logic turned out to be a description of one client's catalogue, which is what separates the engine that gets reused from the configuration that never should be.
-
10 August 2026
The security pass an agent build needs before it earns write access
Scoped actions instead of open API access, why "it's just logging" doesn't stay true, documenting accepted risk instead of leaving it invisible, and why the pass belongs before you widen the surface, not after.
-
1 August 2026
Is your AI assistant telling your patients the truth?
An assistant that is wrong is not obviously wrong. The failure classes that show up months after go-live, why a demo cannot surface them, and how to find out what yours has actually been telling people.
-
29 July 2026
Keeping Plato and Google Calendar in sync, without either one drifting
Two ledgers, one truth. Picking an owner and meaning it, the discriminator problem nobody warns you about, building for the second webhook, and writing the minimum a patient can be identified by.
-
24 July 2026
Augment, don't replace: where AI should act, and where it shouldn't
The volume-vs-judgment split that actually decides where AI should act alone, why the boundary has to be an explicit rule, and how to make a handoff feel like an upgrade instead of a dead end.
-
15 July 2026
Log everything: audit trails for production AI agents
Why a transcript isn't an audit trail, what a log record needs to hold, and how logging every answer and action is what makes accountability real rather than a slogan.
-
1 July 2026
From answering to acting: connecting an AI assistant to live systems
The jump from an assistant that answers to one that books, reschedules, and cancels on live systems: source-of-truth discipline, scoped authority, and safe, logged actions.
-
24 June 2026
Compliant by design: keeping AI answers aligned with AIHGle 2.0
Using agentic methods to keep AI answers compliant — extending the "built to be trusted" line straight from the medical-group case study.
More posts to follow, including a write-up of the outreach build once it's further along.