HEADCOUNT & CODE · A Building Tech Teams publication
01 · THE PATTERN
A common theme across the articles I read and podcast conversations I’ve had lately is that companies want to become AI-native, but most don’t really know what that means or how to measure whether it’s working.
Meta tried to transform teams across one of the biggest technology companies in the world and pulled back. Me&u started with three engineers on one problem, then gave the rest of the company time to learn the tools. At the same time, more engineers are looking at what small teams can now build and considering leaving their jobs to do it themselves.
Building an AI-native startup from the beginning is significantly easier than changing an existing organisation. Jack Rudenko, this week’s guest on the podcast and one of the best engineers I know, said “Ideally, yeah, like kill everything if you can and build it from scratch.” The technology is only one part, though. You also need to change how people work, what they’re measured on, what gets shipped, and be clear about what the change means for their jobs.
The companies making progress appear to be starting smaller, giving the right people meaningful problems and measuring what reaches the customer. The ones that don’t risk producing more work without more value, annoying their employees, or losing their best engineers to companies that will let them build.
02 · THE MOVES
1. Meta proves becoming AI-Native is not so easy.
Meta spent the first half of the year on a plan codenamed Project OT, short for Organization Transformation. AI would take over much of the daily work of thousands of staff, overseen by smaller "talent-dense" groups of humans. Scenario exercises looked at cutting some teams by as much as 60 per cent, in two waves. Hours before the first wave Zuckerberg pulled out. Meta cut 10 per cent the next day and cancelled the second wave. ((Reuters via BNN Bloomberg, 26 August)
Meta has pushed back on some of the framing and numbers, but this is the perfect example of the hype and mostly unrealistic talk of established companies suddenly becoming AI-native. This is one of the best-resourced technology companies in the world, with access to almost unlimited money, talent and computing power, and they are struggling with it.
Meta’s numbers captured the problem. Code changes to its platforms increased by 220% year on year, while changes reaching users increased by 36%. More code was produced, but the improvement reaching the customer was nowhere near the same. Are companies becoming more productive or simply faster at producing more work?
This last point has come up in most of my podcasts lately. Claudia Barriga-Larriviere asked, “Yes, we’re becoming faster, but are we becoming better? Where’s the value?” Adam Witanowski raised the same question through ROI and whether companies are mapping increased AI spend to what the customer actually receives.
The final aspect of the article is the most important for leaders. Meta reportedly put tracking software on US employees’ devices to capture keystrokes and mouse movements so its agents could learn how people worked. When employees got a whiff that the company was teaching AI how to replace them, internal sentiment fell from 74% to 55%.
For leaders: Change management is going to become a much bigger part of what companies expect from technology leaders. If your AI effort is about increasing performance, say that. If it is about reducing headcount, say that instead. Your people are thinking about it either way.
For tech professionals: This is the worst AI will ever be. The tools will improve and companies will keep pursuing better performance, lower costs and increased efficiency. The people succeeding are the ones combining their own knowledge with the tools. Human plus AI is currently better than either one working alone.
2. Me&u started with three people and one real problem.
Me&u, the Sydney hospitality technology company behind the QR ordering system you’ve probably used at a pub, launched a reservations platform last month. CEO Kim Teo says AI accelerated the build, but how they approached it is more interesting than another company claiming it used AI to build something faster. (SmartCompany, 26 August)
It started with three engineers on a stealth project. Teo said they didn’t want to build a small module, widget or customer service automation. They wanted to build something that could become as large as Me&u’s core ordering and payments product.
A small team, a real problem and a brief big enough to be worth their time. Rather than telling the entire organisation it was now AI-native, Me&u gave three engineers somewhere meaningful to start.
Then in April, leadership paused non-essential engineering and product design for a month so the wider company could learn and experiment with the tools. This connects directly to last week’s newsletter and John Collison talking about having to “deprogram” Stripe employees from over-engineered ways of building products.
I’ve heard of a number of companies using a similar playbook. Small team, prove value, then train a wider team. I’ve seen that playbook work significantly more times than the Meta-style approach in the example above.
For leaders: Don’t begin with a company-wide announcement that you’re becoming AI-native. Find the right people, give them a real problem and remove some of their existing workload so they can take it seriously. Measure what happens before trying to transform the whole organisation.
For tech professionals: There is a significant opportunity for people who can see the bigger picture and change what their organisation believes is possible. These are the people getting the largest pay increases at the moment, not the ones using AI to produce the same work slightly faster.
3. Your best engineers want somewhere to build.
Information, media and telecommunications businesses in Australia grew by 4% last financial year, compared with 3.1% growth across all businesses. The founders interviewed by Capital Brief believe layoffs and cheaper tools are both contributing. (Capital Brief, 25 August)
“Thanks to layoffs, we’re also seeing more people taking chances on themselves to actually create a startup,” Relevance AI founder Jacky Koh told Capital Brief.
This echoes Cloë Stanbridge’s comments on the podcast. AirTree’s Frontier program gives founders a $250,000 SAFE cheque to leave their jobs and start building. When it opened, it received four or five hundred applications.
Cloë also told me about someone who left a full-time startup role leading a team of thirty because she wanted time to experiment, use the tools and get back to what she loved doing. Her warning was that if good people work for companies that aren’t moving, or are prevented from using particular tools, they will eventually want to step away.
That makes this a retention problem as much as an entrepreneurship story. Your best engineers can see what individuals and small teams can now build. If the only way they can work that way is to leave, some of them will.
There is an important caveat. Building a product is cheaper, but building a business is not. When everyone can build, knowing what to build, understanding the customer and getting someone to pay become more valuable. I believe GTM will be one of the most valuable and in-demand roles over the next eighteen months.
For leaders: Create opportunities for your best people to build inside your company and in front of customers. Ask what they want before they resign, then give them the environment to show their initiative and entrepreneurialism.
For tech professionals: More than 14,000 Australian companies entered external administration in each of the past two financial years. Going out on your own is not as easy as the fake Instagram business profiles make it look. Some great companies will let you build ambitious things with customers, funding, a salary and a team behind you. Starting your own company is not always the holy grail.
03 · FROM THE INDEX
One read a week from the AI Jobs Index, the live tracker I run at aijobsindex.com.au.
The Index is two weeks away from having a significant uplift. Backdated data, verified from multiple sources, and the ability to analyse AI’s effect on the Australian job market at a much deeper level. Until then, I don’t expect any company to announce layoffs citing AI as the reason.
04 · WHAT I’M BUILDING
To be honest, I didn't build much this past week. I’ve been on the tools recruiting AI Engineers and Heads of Technology.
But I did have some great conversations with experienced tech professionals about best practices for using agents to build for me. More to come in the following weeks.
05 · PEOPLE MOVES - FAKE CAIOs
Commonwealth agencies were directed to appoint a chief AI officer by 1 July. The number went from 56 on 15 June to 113 by 29 June. (Information Age)
As a hiring story, this is bullshit. There is no way 57 government agencies found, assessed and hired properly qualified AI leaders in two weeks. There aren’t enough qualified people in Australia, and the ones who exist are some of the most sought-after talent in the market.
The agencies were told to appoint existing staff. The person didn’t need to be technically qualified, and there was no additional pay. Most agencies gave an existing executive another title, which may create accountability but does not create AI capability. This is government policy dressed up as pushing forward with AI, with no substance behind it.
For leaders: this is the private sector story too. Appointing a Head of AI out of your existing leadership team is a reasonable first step but it is not a capability. If the person has no additional time, no budget and no extra authority, you've created a reporting line or a title addition, not a function.
For tech professionals: if you're offered one of these, ask what comes with it. Time, budget, headcount, decision rights, and whether the title is reflected in your pay. A CAIO title with none of those attached is a waste of time.
06 · THIS WEEK ON THE PODCAST
Jack Rudenko is Chief AI Officer at 10X Labs, and he’s probably the most technical AI engineer I know. He hasn’t written a line of code himself in nine months, and his answer to my first question set the tone: “As always, the biggest bottleneck is humans.”
His view is that most software delivery processes were designed around human weaknesses. Tickets, handovers, code reviews and release gates exist because people forget things, make mistakes and need to communicate. Putting faster agents into the same process doesn’t automatically make the process faster.
The part that stopped me was his answer when I asked how a company with twenty engineers introduces AI properly. He said that answer would make him a billionaire. Most tools and approaches are designed around one person, and he hasn’t seen anyone solve them properly at a team level.
That might be the biggest gap in AI engineering right now. The tools are improving quickly for individuals, but most companies still don’t know how to rebuild the team around them.
This is an episode for anyone who wants to understand the thoughts behind a top-level AI Engineer, what’s possible now and where it could be going.
Listen: YouTube · Apple · Spotify
Have a great week!
Cheers, James
Headcount & Code is the weekly read from Building Tech Teams. Written by James MacDonald, MD of NTP Talent. The AI Jobs Index is a Building Tech Teams publication.