उलझे नहीं तो कैसे सुलझोगे, बिगड़े नहीं तो कैसे संभलोगे If you never get tangled, how will you learn to untangle? If things never go wrong, how will you learn to set them right? I was part of three AI trials inside a New Zealand government agency. All three gave us invaluable lessons. Out of those three, one progressed quite a lot, through four stages of development, and has shown genuine promise. ...
Your organisation has a document management system. Maybe it’s called Objective. Maybe it’s TRIM. Maybe it’s a home-grown records platform that nobody wants to touch because the person who built it left in 2014. Whatever it’s called, it holds years of institutional knowledge. Briefings, reports, decisions, correspondence, approvals. The kind of material that makes new work possible — if you can find it. You can’t. Not quickly, anyway. And that’s the problem this post is about. ...
Most companies don’t fail at AI because of bad models. They fail because they never had a strategy. That might sound blunt, but it’s what shows up again and again in enterprise settings. Teams jump into pilots, experiment with tools, hire a few data scientists—and then stall. Not because the tech didn’t work, but because the system around it didn’t exist. If you’re leading AI inside an organisation, you don’t need more tools. You need structure. ...
Your enterprise already has many channels. Humans use some. Machines use some. And now AI is asking for its own. In the last post, I explained why APIs are not enough and how MCP creates a new AI channel. If you missed that, read it here first: From APIs to AI Channels: The Next Evolution This post is the next step. We zoom out and see where this MCP channel actually fits. ...
Your system already has APIs. Documentation. Schemas. And yet an AI agent still can’t use it properly. That’s not a tooling problem. It’s a channel problem. For decades, every meaningful leap in software came from improving how we communicate with systems. It started simply. You’d walk over to someone’s desk. Ask how things worked. Knowledge lived in people’s heads. Fast if you knew who to ask. Painful if you didn’t. ...
After assembling the robot and working through the debugging in Part 2, the moment finally came. Power on. Motors twitch. Antennas move. And then… a small cheerful “phew phew” whistle. Reachy didn’t speak words. But in that moment it absolutely felt like a greeting. Kia ora. The First Boot — Kia Ora Aotearoa The first time Reachy powered up properly it made a small whistle-like sound while its motors calibrated. It isn’t speech. It’s just a sound effect the robot makes as it wakes up. ...
Vibe coding is fun. It’s fast. It feels like magic. But unmanaged magic in a production environment? That’s how you get security gaps, compliance headaches, and a quiet email from the risk team asking for a “quick chat”. There’s a better way. For complimentary app development — dashboards, internal tools, workflow helpers, automation layers — managed vibe coding can dramatically increase output without increasing exposure. The trick is simple: structure the chaos. ...
Back to Part 1 — Reachy Mini Arrives in Aotearoa Sometimes a robot just refuses to cooperate. After spending seven months waiting for my Reachy Mini to arrive, I assembled it with care — and then watched it refuse to boot. Power LED: green. ACT LED: one sad blink, then nothing. No WiFi hotspot. USB-C ignored. The motors twitched promisingly, but that was it. My little desk robot was alive enough to taunt me, but too broken to respond. ...
There are two kinds of tech excitement. The day you order something… and the day it finally arrives. Between those two moments lives a strange ritual: refreshing tracking pages, checking emails, and convincing yourself the package has definitely moved since the last refresh. My Reachy Mini robot took seven months to travel from idea to desk. This post is Part 1 of a multipart story. Reachy Mini Diaries — The Kiwi Robotics Journey Part 1 — Waiting, Arrival, Assembly (this post) Part 2 — Debugging the Robot Part 3 — First Impressions Part 4 onwards — Building Things With Reachy The Order (July 2025 Optimism) In July 2025 I ordered a Reachy Mini, a small open-source desk robot created by Pollen Robotics in collaboration with Hugging Face. ...
For most of modern software history, documentation existed for one core reason: human written code was expensive to create and difficult to replace. Developers carried complex mental models in their heads. Code was the physical imprint of those models. Documentation was the act of extracting that fragile, human knowledge and making it reusable. Documentation existed to protect human intellectual property. That logic is breaking. With systems like OpenAI Codex and other AI coding agents producing usable code in minutes, the implementation itself is no longer the scarce asset. Code can be regenerated, rewritten, or discarded with very little cost. What remains hard, slow, and deeply human is deciding why a system exists, what it must guarantee, and where its limits lie. ...