HEADCOUNT & CODE · A Building Tech Teams publication
01 · THE PATTERN
Three stories again this week (like every week), and the pattern this week: adaptation.
Life360 cut roles in April and named AI as the reason. Four months later, it’s hiring again, and the roles confirm what they said in April: they were “making a strategic shift toward an AI-native operational model”. Marty Cagan presents a strong argument for the ongoing need for Product people. When a lot of people are talking about engineers wiping out the need for product people, the article presents the case for core product skills that are often rare. Coinbase has spent a year rebuilding its engineering interviews because their interviews were previously designed for traditional software engineering practices.
Three different stories, and in a time where things are changing so quickly, all three point to one of the most important skills right now: the ability to adapt. Adapt to headcount and role scope, adapt to the make-up of a quality software team, adapt to how you hire the best engineers.
I’ve been speaking with Senior AI Engineers all week for a couple of roles I’m recruiting. The best ones understand the need to move, learn (adapt) and stay relevant with the latest technology. But I’ll caveat that: the best senior engineers, even those building their own harnesses and architecting AI solutions, have strong foundations in software engineering. So whilst learning the tools and leaning into AI is relevant, strong software foundations are the key.
02 · THE MOVES
1. Life360 cut for AI in April. It is now hiring AI Native.
Chief executive Lauren Antonoff told Capital Brief last Tuesday that headcount growth is back on. “We still expect that over time, headcount growth will continue,” she said, and she explained what the April cuts had been for. “We just reduced our headcount a bit, so that we had more flexibility for hiring and to figure out what we needed as we move into this world.” (Capital Brief, 11 August)
The most interesting part of watching this story play out is seeing the roles Life360 is hiring back into its team now that it has that flexibility, and how it's following through on that shift toward an AI-native operating model.
Life360 has 28 roles advertised. Nine of them sit under Engineering, and eight of those nine have “AI Native” written into the job title.
This is a story of a company that looked at its operating model, saw an opportunity to make a significant change, went through redundancies, and is now actively moving to a new team makeup. They are investing heavily in AI, and the engineering team they are hiring for is built around real AI capability. When companies start building this properly, it starts with Data, Machine Learning, and analytics. They are not simply hiring software engineers using Cursor.
For leaders: I’m seeing two ways companies are increasing AI capability within their teams. 1) Many companies are folding AI skills into current roles, and we’re now seeing AI mentioned in 34% of tech job advertisements. 2) If your company wants to make a serious play in AI, the AI capability needs to be built from the ground up, with AI-specific roles and responsibilities, often including a mix of architects, data engineers, ML engineers, and senior software engineers.
For tech professionals: When “AI Native” is in the actual job title, rather than a mention in the job advertisement, the interview is going to test you in depth; you cannot call yourself an AI Solution Architect if you’re an Architect who uses LLMs. I’ve seen this firsthand over the past week; we’ve had applicants for Senior AI Engineer roles who are building their own harnesses versus engineers using Copilot - the difference is significant.
2. The product role is essential.
Product Managers have had as many questions asked about the future of their role as nearly anyone. Is the future Product Managers who can now build? Or Engineers who learn product? Are the best teams of the future a combination of engineering specialists and product specialists as stand-alones? These questions comes up nearly every week in recruitment discussions.
Marty Cagan published a piece last Monday, quoting analyst Benedict Evans, who argues that giving everyone the ability to build tools does not turn everyone into a tool builder. Evans describes three skills, and Cagan’s point is that they add up to the clearest description of a strong product person he has read. (SVPG, 10 August)
Characteristics of the strong Product Person:
Lots of people recognise pain, and maybe have ideas for addressing that pain, but not everyone is able to see the more general problem to be solved behind that pain or idea.
Being able to find a solution that actually works: “People that are really good at using the tool are not the same people as those that are really good at creating the tool.”
“Beyond discovering a strong solution that meets the needs of your customer, you also need to have the depth and breadth of understanding of your company to discover a solution that also works for your business”.
Evans’s closing point confirms something I have written about before in this newsletter: the element of “taste”. Just because everyone can build doesn’t mean everyone should build.
For leaders: great engineers don’t necessarily make great product people, and likewise great product people don’t make great engineers just because they have tools to help them write code.
For tech professionals: if you have been hearing that AI makes the product role redundant, this is a strong argument against it. The skills that are hard to copy are problem selection and judgment (taste). Average product people may be at risk, but the best product people are the difference between building slop that doesn’t fulfil customer needs and a product that is successful.
3. Coinbase spent a year rebuilding its engineering interview because the old one was testing a job nobody does anymore.
AI-generated code went from 5.7 per cent of everything merged into Coinbase’s codebase in the first quarter of 2025, past 50 per cent in the fourth quarter, to roughly all of it now, with humans still reviewing every line. (Coinbase, 13 July)
Their own description of what changed:
“It changed how our engineers spend their time. Less writing code from scratch, more directing AI, reviewing its output, catching its architectural mistakes, and making judgment calls that models still can’t make.”
“When the cost of building goes to zero, the cost of identifying what to build, verifying it’s correct, and getting it out safely becomes the limiting factor.”
Back in Issue #5, I wrote about the bottleneck moving from writing the code to reviewing it, after Gergely Orosz reported that the code review pile-up was the fastest-rising worry among the engineering leaders he speaks to.
So they rebuilt the interview process around what their engineers actually do day-to-day:
The new frontend interview. Candidates are assessed on the quality of their prompts, how they evaluate the output, how they catch errors, and how they iterate.
The new backend interview. Not writing from scratch, but working with and against an existing codebase and an AI collaborator to produce an improved result. Can they triage real issues? Can they catch subtle bugs that AI introduces?
And what they actually score:
Usage. Does the candidate use AI tools effectively and responsibly? Do they select the right tool, apply it in the right parts of their workflow, and produce measurably better outcomes?
Application. Do they know when AI is and isn’t the right solution? Can they design AI-enabled workflows that create real business impact rather than just automating tasks?
Understanding Limits. Do they understand where AI breaks down? Can they identify the privacy and security implications, and apply appropriate human judgment as a guardrail?
This is a real trend we are going to keep seeing with interviews changing, because when you have people like me who can build (and I use that word lightly), there is now a big difference between what I am and what a quality software engineer is. Making sure your interview tests for engineers who can produce high-quality code, understand what code has actually been produced, and test and bug-fix that code is arguably more essential now than ever - even if the engineer isn’t writing each line.
For leaders: I still believe that the best Engineers, even those not writing a line of code themselves, are the ones with sound software engineering principles. I think leaders should still test for those competencies, but it's worth asking: is your technical interview the same as it was two years ago? Does it match the way your engineers actually work day-to-day?
For tech professionals: Coinbase has published a pattern that I think we’ll see variants of becoming common across the industry. Not only understanding how to prompt and how to use the right tools, but actually pushing back on the models that claim they have the right answers and the right code. Being able to find bugs and loopholes is becoming the real differentiator for quality AI-fluent engineers.
03 · FROM THE INDEX
One read a week from the AI Jobs Index, the live tracker I run at aijobsindex.com.au.
A third of the technology and data-adjacent job ads the Index scans now mention AI. The diffusion index reads 34.41 per cent: 1,348 of 3,918 Australian ads in the trailing month.
What this means: The story has moved from layoffs to the changing nature of roles across technology. AI is being folded into a third of all tech and data roles.
04 · WHAT I’M BUILDING
A software engineering team. Who would have thought that a few months ago!
I will start this section with FEEDBACK WELCOME. I am not a software engineer, but last week’s conversation with Adam Witanowski and his dark factory had me excited, and I was equally as annoyed at my Claude over the weekend.
I built AI Jobs Index a few months ago, and it’s about to get a significant upgrade with higher quality data sources. But the Index also spent five days serving zeros this month before I noticed - an expired key, and no check anywhere that would have told me. And weekly, I'm frustrated that my poorly architected solution keeps breaking. Maybe I should have taken my own advice: “just because everyone can build, doesn’t mean they should build”.
But I enjoy the learning process, so the AI Jobs Index is getting a team. What got me here was enough to get it live, but nowhere near enough to make it reliable. The next version needs someone who can challenge the build decisions, test them, not just execute them. Welcome, my team of agents.
Phase one is complete. Acceptance tests are built in along the way until I get a system I trust. It’s an absolute work in progress.
05 · THIS WEEK ON THE PODCAST
Matt McFarlane runs FNDN, and he knows compensation in Australian tech and startups better than anyone I speak to.
His opening line set the tone: “the number is only half the story.” Most companies, in his experience, have no rationale behind the figure. “Someone’s negotiated for it or they’ve plucked it out of thin air or a mate in the ecosystem has said that this is what the role’s worth.”
Three parts I found most interesting:
Why AI salaries actually spiked, and it is not the hype. “Payroll’s always been, in SaaS, the biggest cost”. AWS has been up there as well, but token costs and AI infrastructure are being considered now, too. When the infrastructure bill increases, companies are paying the people who understand it. (It was the Canva infrastructure story of two weeks ago!)
Companies are banking the upside. A new hire negotiates above the person already doing the job. Matt’s view is that the incumbent is worth more, not less, because of the context they carry, and that most companies sit on it: “a lot of companies think, oh, let’s just see if we can bank that upside until it becomes an issue. But by that time, the trust’s gone.”
What promotion looks like when the team has agents. He described a CTO who now expects his people to be promoted by building agents that automate the level of the role they are doing. “They’re probably never gonna hire a junior again, which is crazy, and then eventually those, and there’s maybe eight or nine of them, they’ll all just be principal-level roles with a whole bunch of agents under them.”
A key line for managers who are promoting AI within their companies: AI usage is not a performance indicator. “I could spin up a loop right now that just spends tokens and would make me look brilliant despite achieving nothing.” His answer is that we will need a way to judge whether what an agent produces is worth more than its token cost, or the cost of hiring someone to do the same thing.
Listen: YouTube · Apple · Spotify
06 · THE ASK
This week’s podcast guest, Matt, runs Startup People Summit, and it's happening on September 3rd. I’ve booked my ticket and paid full price - this is not a paid advertisement. I just believe this is someone in the People community putting together an event with fantastic speakers discussing topics relevant to Building Tech Teams. Tickets are still available, and I’m encouraging everyone to check out the speakers and topics at startuppeoplesummit.com and get yourself a ticket.
Hope you’re having a great week,
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.