The unhinged advantage

A few years ago, inside a tech company, I asked for a piece of work to be delivered in a week. The answer was: "one week is impossible, it will take one month", because it first had to clear several committees, which meant the whole project would probably slip by three months. I pushed back, offered to sit down with the team and build it together, and we had it done in days. The verdict that travelled back through the building came in one word: "unhinged".

I thought about that word again in early August, when Sequoia's Alfred Lin reshared a memo that Tanay Tandon (CEO of the healthcare software company Commure) once sent his team. Commure's most important value, and the one most often challenged, is "Speed Above All Else". Several times a week, Tandon wrote, someone asks him “if we are in healthcare, how can our value be speed”, or tells him “you can either have it good, or you can have it fast”. He calls that “a false dichotomy meant to pigeon-hole you into mediocrity” and lists his evidence: Bret Taylor rebuilding Google Maps from scratch in a 48-hour weekend binge, the Empire State Building finished in 410 days and still standing, next to San Francisco's Millennium Tower, five years of committees and now tilting. Among the notes he passes on: hire people who are “a little damaged / unhinged”, because intensity rarely comes from normal functioning people.

"Good or fast" is a trap built by people who can't hold rigour and speed at the same time, and the person who refuses the trap is usually the one still looking at the whole picture.

What has changed this year is how short the timelines have become. Michał Nedoszytko, an interventional cardiologist in Brussels, placed third out of 13,000 at Anthropic's hackathon with a clinical tool built in seven days, coded in hospital corridors between cases. On 29 July, at the first Abridge, Anthropic and Lightspeed hackathon on agentic AI in healthcare, 199 participants shipped 113 projects in 12 hours. David Armstrong, professor of surgery at USC, turned 15 years of amputation evidence into three public advocacy platforms in under a week and published the method in Surgery this month with “vibe coding” in the title. Among the newest entries in the directory of clinician-builders I curate is Giftneville Onyango, a newly registered nurse in Eldoret, Kenya: three browser-based clinical tools, built solo, in three days.

And the tools are now accelerating the accelerators. On 13 August OpenAI previewed an "Ultrafast" tier that runs its frontier model up to 14 times faster; its own researchers now fit several experiment loops into a working day where they used to launch one batch overnight. That is a flywheel: whatever already moved fast moves faster, each loop closes sooner, so you learn sooner and start the next one sooner. The announcement's subtitle makes the pitch explicit: “turning speed into a competitive advantage”. On 21 August OpenAI also cut that model's API pricing by over 20% for three months; the flywheel is getting cheaper as it accelerates.

There is a fair objection, and this month it carries the memo's own logo. On 12 August, eight days after Lin's post, STAT published a months-long investigation into Commure's “mad dash to automate the business of health care”, its phrase: thousands of dollars offered to customers and partners who refer its AI products, some clinics reporting steep financial losses, and, per Digital Health Wire's summary, a Wyoming rehab clinic $500,000 behind on collections and an Arizona health centre watching $1 million in dental claims get denied without a single notification. Commure called the referral practice industry standard and the complaints a small minority, and has since terminated the payments; the notices went out six days before the story ran. An essay called The Religion of Speed, popular on Hacker News this summer, gives the charge its general form: “a lot of what gets called speed is just impatience with a different name”.

Both land, and neither touches the argument, because the memo draws the line itself. The sprint, Tandon writes, comes with edge-case detection, testing frameworks, pre-mortems and heads-down focus; what STAT describes looks like a sales engine that outran them. Speed with the discipline packed in is the advantage; speed without it is haste, which the memo's own condition rules out: speed must not come at the cost of rigour. Committees rarely add that rigour either. They mostly add the feeling of safety, without anyone having to understand the work. PwC's survey of 1,004 financial-services executives, published on 3 August, shows where that leaves most organisations: 90% say their industry needs to get more comfortable moving quickly, 70% say they're already moving faster, and 77% still say they're not moving fast enough to keep pace with AI. Even the fast ones feel late.

For healthcare, the current version is this: the learning loop can run at weekend speed while the deployment gate stays as strict as it needs to be. Prototype in days and validate properly; the mistake is treating those as one step. I closed my talk in Berlin in June with Ash Maurya's line, “speed of learning is the new unfair advantage”, and it's the bet behind every project I run. It's also, I suspect, why the word unhinged gets thrown around at all: an organisation that moves at committee pace reads genuine urgency as "instability", when most of the time it's "clarity".

So here is the experiment. Take one thing your organisation has scheduled for next quarter and build a one-week version of it this week, tests and pre-mortem included. Then look hard at what the committee was protecting.

💥 May this inspire you to find your unhinged people and put a testing framework around them instead of a committee in front of them.