My biggest takeaways from my interview with Steve Hind, who grew Lorikeet through 2026 with virtually no new headcount.
AI-native is a different lens on problems, where the challenge is to “define packages of work, hand them off to AI agents and have the agents do the work on their behalf.” Consequently each team member’s capacity for work is unbounded as it no longer is constrained by hours in the day.
Lorikeet’s business growth in 2026 has been rapid, but their headcount growth hasn’t. Steve assigns this remarkable feat to the increasing use of AI.
The first challenge of becoming AI-native was engineers adjusting from writing code by hand to a world where they’re doing “the meta-engineering of figuring out how to build a system that was capable of self-verification and testing.” But once they got over the enormous shock of their job changing overnight, they realised it’s just another rung up the abstraction ladder.
The second challenge of becoming AI-native was that coding agents and associated tools helped non-technical people do the work of engineers. This led to the realised engineering isn’t about the knowledge of code and syntax. “It is a way of thinking about building systems to solve problems.” And realise the work is the maintenance of the iteration and that can result in you “drowning in tools that don’t quite work.”
It’s a bigger change in work for non-engineers, as their work has fundamentally changed from doing your job, to building systems to do your job for you. While engineers have moved up the abstraction ladder, the outcome of their job has essentially remained the same.
Steve’s view on AI and automation has shifted from “letting 1000 flowers bloom” to getting builders to build. “You are probably better off having a minority of people building and pushing tools out for everyone to use than to have everyone building an overlapping, not quite interlocking, not quite harmonious sets of tools.” and “operational teams to have someone on their team primarily charged with systems and automation, and then to have the remaining people on the team doing work aided by those systems and automation”
“It’s easier to change an agent than people.” Steve still believes in training people to do good work, but to get good outcomes he also now believes good agents that follow best practice is a quicker path. “it’s a faster process to push a good update and have them think about it… rather than a very lossy process of a lengthy change management to explain new skills to people.”
A killer combo is pairing builders with subject matter experts. Steve has found that mixing those two people means that they can move faster than, say, trying to turn a business person into an engineer if they don’t have that mindset already.
Having a central business function for agents helps people focus on the specifics of their problem, rather than to waste time on how to build agents generally. “Having infrastructure and blessed patterns for building agents is a central pursuit.” “You want as much of the time people spend to be focused on the unique element of their task and not anything generic to the task.”
Lorikeet are not building internal AI agent infrastructure joyfully. While they have constructed their internal knowledge layer for agents to get the context they need, he feels that over time there will be vendorisation of these tool - they just haven’t caught up yet. He’s cautiously making tools on open platforms like GitHub, using Markdown documents, with a sprinkling of Notion on top, rather than using proprietary platforms like Claude Cloud Routines which might be a dead end for them in the future should they leave the Claude ecosystem.
Steve thinks that you shouldn’t automate your role, you should go and find the things you don’t have time to do and automate that instead. Because the amount you can automate in your job is bounded, but the amount you can automate which you don’t have time for is unbounded.
Rapid fire awesome automations:
Perfect answers to objections scraped from sales calls and shared as training material\
Sales battle cards - instead writing them out for sales in advance, they now generate them quicker/better on the fly using facts stored in their knowledge base.
Customer feedback on features automatically found and digested
“Grinding as a service” what have you done in previous roles which were successful but required unsustainable levels of effort?
AI still grinds Steve’s gears. It still sucks at writing, which makes it bad at interpersonal communication. “this is the crux” urgh. And it has a “polyannerish tendency to be like a bit overly dramatic about things.”
Since January he’s gotten all team members on Claude Code as “the space to like dent the car is greater, but like the space for really high output is massively bigger.” But he’s not worries about dinging up the car because “the capacity to learn now is just like so, so much better and so therefore I think we should, be willing to believe in people a bit more and put more powerful tools in their hands.”
AI gives Lorikeet a third way to improve team effectiveness beyond training and compliance - the agent operating system. So when his CEO daily planning skill was deleted, he thought about how do we improve systems so it doesn’t happen again.
Steve views AI adoption as best being achieved by getting the early 30th percentile started and then raising expectations for the rest. This encourages people to cross the chasm to keep up with raised expectations, rather than being managed out for not using a tool.
After the AI roles apocalypse, Steve’s thinks the only roles left are engineer and salesperson (and maybe owner). Engineers builds a scalable system to solve a problem, sellers convert interest in the market into demand by building relationships, owners have responsibility for performance and compliance. A finance team could just be a governance owner working alongside an engineer who builds the systems, rather than ops people running the function.
The rise of builders. With AI, Steve focuses more on hiring people who show evidence of building and learning. Those who’s “inclination is towards building the system to solve the problem is what you want because they’re the ones who you would think are going to be able to repeatedly build and deploy and manage agents to do more and more work.”
The future of work is less work. Steve sees a future where the massive increase of productivity AI causes allows us to spend more time preparing for work, or increased retirements. This continues a trend which productivity has enabled. But he remains optimistic that capitalism will steer us away from the more dystopian AI doom scenarios.