We don't hire seniors anymore

Juniors + AI = talent arbitrage | Why pre-AI CEOs should cut deeper | A case study for AI change

My approach to hiring has shifted fundamentally in the move from startup #1 (pre-LLM) to startup #2, an Opus Baby. We are taking a bet on hiring inexperienced, low-ego, fast learners with no priors, rather than seniors with deeper domain expertise. Strategy is interesting because it’s falsifiable: one company should be able to disagree with your strategy and take a different and equally legitimate tack. I know founders who are doing the opposite to this, reducing junior roles and maximising senior leverage with AI. But here’s my take.

Startups shouldn’t hire juniors - until now

At my first startup, we hired for experience. Even the sharpest junior hires struggled. They lacked sufficient structure and senior guidance, especially when we went remote during Covid. I’ve interviewed hundreds of early-career, ambitious people who told me they were looking for a new role because the structure, mentoring, or career ladder they were promised by another startup did not exist. This was a red flag for me. Nothing wrong with them, just something endemic to startups, and we too were a startup.

Startups at their peak are like SAS units: small, fast, efficient, violent. There is not space within that for much structure, training, guidance. I’m sure counter-examples exist, but I think I speak for the majority. By contrast, large companies are like a whole army: they have a training corps with its own staff, a graduation process to full roles, a ladder for career progression. They’re effective at scale but they do not move quickly. When I was won over by a charismatic junior person and hired them, it rarely worked out. And for no fault of their own; I had failed to acknowledge our mutual incompatibility.

Now you should only hire juniors?

Now everything has changed, because across many skills, senior-level knowledge and skill have moved from human brains to machine ones. At the same time, absolutely everything has changed so previous experience and knowledge are not as valuable as they once were. This is not a blanket rule, but it applies to many, many role categories.

The talent shift this is precipitating can be felt most keenly when you are a brand new startup. We at Rig have no cultural priors to shift, no technical debt to work through, no workforce to upskill or coax or re-engage. Our first hires had 0 & 2 years of work experience respectively. Almost no priors from the pre-AI era, low-ego, and a massive appetite to learn and succeed in AI. From day 1, they lived inside Claude Code or Cowork. Much of the building & operating work we do outside those tools now feels inefficient, or a problem to solve. Do we have to click around a UI? Could we automate x, y and z?

This will drive a massive change in what it means to be senior. The term will cease to be a recognition of knowledge or time spent at an organisation. It will become a recognition of endurance, context-switching capacity, speed of learning, and taste. The shape of seniority will be completely redefined by the arrival of AI in organisations. A senior will be the person who spikes highest on 2-4 of those factors.

Tomorrow’s ‘Builder’, as the role is coming to be known, can be measured across two axes:

• Endurance and context switching are measures of someone’s capacity to take on multiple roles in parallel. You can talk to customers, analyse data, prototype (including designing *nice enough* things), build, and ship. You can market it too. This is a lot of work for one person. It would take a team of 7 in the past. But it IS possible. It is hard, and tiring, but the only constraints to doing it now are endurance and context switching. It looks like someone staring at a black screen all day. They’re running a lot of agent tabs in parallel, without losing energy or focus.

• Speed of learning and taste is the other axis. To both keep up with the changes in AI, and to leverage tools to address real customer problems in a way that delights and satisfies them requires both. AI is moving so fast that almost every aspect of work has a monthly frontier shift. Coming back to the army analogy, moving an entire front line is a slow operation and opens many gaps that cause losses and delays. An SAS unit can move its full frontline in minutes, because it is a small group with high endurance and speed to adapt. But only when they can learn and move fast, and apply judgement, pragmatism, and taste to each situation very fast.

We may hire ‘seniors’ in the old-fashioned sense of ‘those with more years of experience’, but to pay the premium for them, they must look like tomorrow’s seniors, and apply their years of knowledge-building on top. But for many, they are trapped by their prior way of doing things, or their expectations of what work looks and feels like. The ZIRP/Covid era has spoiled many of us in this regard. Perhaps we had forgotten that we work for organisations, not the other way round.

This all assumes you are in an organisation that *allows* employees to access the best tools - and here too, the frontier is moving. Right now, at time of writing, you need to be allowed to use Claude Code or Claude Cowork to maximise your AI capabilities. You need to be allowed to maximise token usage on the best models. You mustn’t be locked into Copilot, like all my non-startup friends are in corporate life. And - forgive my plug - you really need to be allowed to use tools like Rig, so your Claude chats can reliably and safely access and work with internal data. The reality of this is gradually going to hit corporate IT and Finance teams, unless Microsoft catch up extremely fast. A startup’s advantage over an incumbent - until it does - is massive.

The result of this flattening of the experience curve is that a recent graduate can become ‘senior’ much faster. If you have high speed of learning and intellectual capacity, you can keep up with the frontier and understand the complexities of your own organisation’s product. If you build the intangible people and communication skills you need alongside your capacity to build, you can leapfrog people many years ahead of you in their career. This lets you sell and drive change, internally and externally. If you build endurance for sprint and long distance, and develop the capacity to context switch, you can manage a fleet of agents executing work across projects in parallel.

If you are a strong communicator, a salesperson, and generally apolitical, you are feature complete. This is the autonomous senior builder. Be this person!

As a selfish aside - if you are already becoming this person, please get in touch with me - we’d love to meet you at Rig.

A case study of change

At Rig we work closely with Cleo, a $350mn ARR fintech with c. 500 employees. Cleo began working with AI a long time ago. It was a chatbot in 2015, long before transformers and BERT emerged, let alone GPT-anything. But things have changed.

Realising the need to drive a talent shift to AI-first Builders, Barney, the CEO, recently changed the budgetary policy on AI: out was the corporate ChatGPT license policy, out were token caps. In is widespread access to Claude AND ChatGPT; in is ‘tokenmaxxing’. In is several days where the whole organisation downs tools, and picks up new ones: building days, dedicated to just building whatever you want with AI tools. The data team, with whom I work most closely, ran a day for the team training on building agents and adopting MCPs inside Cursor and Claude, using Rig MCP to connect up to internal data, and Notion MCP for internal knowledge.

But a day is not enough: with 500 people, the anti-gravitational force required to drive behavioural change in existing employees is huge; prior ways of doing things are deeply rooted for all of us. The next steps will involve happy chaos: multiple internal AI projects, many built ad discarded for the sake of learning, some kept for the sake of the business. It will be productive for the business, but the arrow here is not straight, because it’s making a genetic change that grows the topline later, not tomorrow.

What should Barney’s exec team do next? Unplug the mouse for their team. Force teams to work from Claude or Cursor for a day on jobs that are non-obvious. Let them find the edges, work around the gaps, discover the possibilities by being forced to throw off their priors.

CEOs need to cut deeper

Any large org that wants to become truly AI-native will also need to cut deeper. CEOs right now are quietly allowing natural churn to winnow their organisation. It’s post-Covid ‘right-sizing’. But some are rightfully making active cuts. This sounds harsh, but I think it can serve all sides. On the business side, in order to pivot into an AI-native organisation, a CEO and leadership team need to change their DNA. It means a team of majority AI-first builders, whether new hires or re-skilled existing team. It means a small group of leaders who are living, every single day, inside the latest breed of AI tools, and have the space - clear of HR overhead and calendar chaos - to use them. It means a much flatter org, with no middle managers whose jobs are hosting meetings about meetings. It means opening up to flexible AI tool use, letting people buy from more vendors, and try many different types of AI tool in parallel. The signal to noise ratio of the org, and its attitude to risk and expenditure on tools has to change very fast to create the space for change.

On the talent side, anyone working in an organisation that is not making this dramatic shift is in a terrible and losing position: stalemate.

Stalemate is much worse than losing, because you stay there forever. You are in an org that does not give you the tools or time to rebuild your own skillset and mental priors from the ground up. And by being in the org, you are part of the people-shaped bottleneck to rapid change. Being made redundant is horrible and I do not think every organisation should cut headcount nor needs to. But being ambitious, and stuck in a role where you cannot reinvent your skillset for the fundamentally different technology age that we now live in, is dragging you behind your peers for the next role at terrifying speed. If you are not learning to master AI tools right now, the next role will get harder to land on a log scale, not a linear one, because other people already are.

What should you do if you’re in that position? You can start building. Use tools at the edge: learn to navigate a terminal and use Claude Code, not the easy path of Lovable or Replit. You can build side projects in life, or shadow projects at work. Maybe you get fired, but maybe you become the change. You can quit, and join an AI-native organisation. Take the pay cut - you’ll get skill training by being forced to use those tools and work faster than you can imagine. If you’re an executive, you can accept risk as a policy: open the vendor floodgate. Rely on your team to use tools wisely. Accept the trade-off of a data-leak risk from an employee uploading a CSV to Claude vs the risk of being destroyed by competition. Irrelevance or collapse are surely worse.