# Show Me Your AI with Danbot Gilbot — How We Hire for AI-Native Talent

*Guest: Alice B.*

---

**Dan:** Welcome to Show Me Your AI with Danbot Gilbot. Any company trying to go AI-native wants to hire for AI-native talent. They've probably already seen that someone who uses AI well can 10x the output of a normal employee. We actually changed entirely how we hire — not just the job postings, but the criteria, what we look for in a person, how we interview them, what "qualified" looks like when the tools are changing every few months. So with me I've got Alice B.

**Alice:** Hey Dan. Danbot Gilbot.

**Dan:** Walk us through it. What's actually different now?

## The problem: AI-native unicorns don't exist

**Alice:** First things first, the talent that we're all dreaming of — that has years and years of experience in AI — just doesn't exist. So let's all stop chasing unicorns. You can see the tweet here: the company looking for 10 years of hands-on experience in generative AI models. It reminds me of *The Mathematics of Love* by Hannah Fry, where she talks about finding an ideal partner — and with some basic criteria, suddenly you've ruled out half of the world.

**Dan:** Yeah, this is quite literally an impossible ask. But even if we were saying we want somebody with incredible industry and client experience who has used Claude regularly, we are already looking at a really small pool of people and losing some potentially great talent. So if we think about a supply and demand curve here — the needs and wants we have from future candidates are changing rapidly, but the supply of candidates with experience in those areas just isn't growing at the same rate. So we need to be hiring for potential and then developing skills through training once people join, in order to plug that skill gap.

**Alice:** This is even more true for us, because we're trying to push the boundaries of what's possible with AI infrastructure. And when we're interviewing candidates from elsewhere, they've not even been given a token allowance, let alone access to some of the tools.

**Dan:** We shouldn't be ruling out candidates because their previous companies weren't embracing AI as much as us.

**Alice:** And of course, our demand curve is steeper than perhaps other companies right now.

## Step 1: Cognitive assessments

**Dan:** We've taken this as an opportunity to review the criteria we're looking at for the ideal candidate to join Brainlabs. We had pretty strong criteria already — in some cases, unicorn criteria.

**Alice:** We were looking for people that had amazing baseline industry experience, strong values alignment with our culture code, a growth mindset, and a really high level of cognitive strength as well. And if you're not looking for that last criteria already and you are an AI-forward company, then I really strongly recommend introducing cognitive assessments into your hiring program.

**Dan:** We do that with Criteria.

**Alice:** They don't call it an IQ test, but it's spatial reasoning, verbal ability, and also maths and logic. They've done loads of research that shows your combined score across all of those three positively correlates with performance in the majority of industries and roles. So regardless of the role someone's joining, every single candidate that goes through our process completes these tests.

**Dan:** But even more so, you say, for AI readiness.

**Alice:** Yeah. If this is not something you're doing already, I strongly recommend introducing it.

## Step 2: Screen for AI adoption (Rogers' diffusion model)

**Dan:** That's step one. The next is extensive AI adoption from anybody that's going to join the company — Rogers' diffusion model, which applies to almost all new technology and how quickly people adopt it. Let's run through these categories.

**Alice:** You have different people at different stages. Your laggards are people saying the whole thing is going to burst — won't touch AI.

**Dan:** "AI sucks."

**Alice:** Yeah, "it's ruining the environment." And on the other side of the scale, we've got innovators — people who are marbling it through everything they do.

**Dan:** And this would be outside of work, to your earlier point. These might be people uploading their genetic sequence plus their health data and bloodwork, training it on doctors, then using that to analyze their Whoop data or something.

**Alice:** Yeah, people that are really leading the way. I've done that.

**Dan:** Well done. Thank you.

**Alice:** The early majority would be people who right now are spending a lot of time in ChatGPT but using it as a search engine and not much more. As Dan said, we can't expect people to have had a huge breadth of experience in their workplaces, but we do expect any candidate that gets far through our process to be an innovator or early adopter in either life or work, and to be able to evidence that. We're using that now as a screening criteria right at the top of our process.

**Dan:** Because if they're not interested in AI, they're not going to fit well in an AI-native company.

**Alice:** And they're easy questions to ask — we're asking which LLM they're choosing to use. So anybody running those screening calls can easily filter people out on that basis.

## Step 3: Systems thinking / agentic management test

**Alice:** That's step two. Step three is probably where I want to spend the most time. We really need people who can act as orchestrators of agents. Few people in our industry or beyond will have experience doing that, and managing agents is its own skill that requires different management skills to managing a human.

**Dan:** We're defining this as systems thinking, but very specifically looking at agentic management. I spent a long time looking at how companies assess for this skill set — and honestly, tell me if anyone else is doing it, but I couldn't find anything.

**Alice:** It's new. So we developed something ourselves.

**Dan:** I remember my university entrance exams where they invented a new language and you had to extract from that language to pass the test. They didn't use a known language — otherwise, if it was Greek and someone happened to speak Greek, they'd have passed straight away, an unfair advantage. What you're really doing is setting up a systems thinking / agentic management test logic that does not rely on whether they've worked with agents before, because that could just be a factor of whether they worked in a company that's using agents.

**Alice:** Absolutely. And this is a pen-and-paper task. We're not asking somebody to spend time in a particular LLM, so no prior experience advantages you. I'm not going to actually do the assessment on you, Dan, because I don't want you to give everyone the answers. We'll do it separately, but I'll talk you through the flow, and we'll share more details if you want to use it yourself.

**Dan:** And if anyone applying is smart enough to have watched our videos, they're going to pass that test.

**Alice:** Yeah, you can just tell us you've watched it and we'll skip this step.

**Dan:** We'll pass them for initiative.

### Part 1 — Systems thinking

**Alice:** First up is systems thinking. We lay out an example of where you may use agents to support you, and explain what an agent is for those who haven't used them. There are three different agents: an organizer who manages your calendar requests, an inbox manager who handles emails and sends replies, and a safety filter who monitors the other two for any risky actions. The first step asks the candidate to tell us how these three agents may communicate with each other, and to draw an arrow diagram showing how they'd work. This is us seeing if they can make these agents work together in an efficient and effective way. Once they've done that, we define the correct answer so we're working on the same basis moving forward.

### Part 2 — Something goes wrong

**Dan:** Something goes wrong.

**Alice:** As always happens with any agent experiment. In this scenario, whilst you're on holiday, a salesperson asks for an urgent meeting — and that person isn't actually urgent in reality. The inbox manager reads the email, flags it as urgent, asks the organizer to book a slot, and that organizer pushes back an important client meeting for this apparently urgent meeting. We can all agree that's the incorrect action — a human would not have taken it. We then work with the candidate to work out what went wrong and how those agents could have been better instructed to prevent it in the future.

**Dan:** This is about diagnosing problems and understanding what in your system could change to be improved.

### Part 3 — Instruction clarity

**Alice:** We then move on to instruction clarity. People will have had a different degree of experience, but we're looking for people who can be specific in their prompting and get the best out of agents. We give them some intentionally bad instructions that created this bad impact, and ask them to choose one, rewrite it, and tell us how. A simple, agnostic way to see how people prompt an agent.

### Part 4 — Human in the loop

**Alice:** For section four we talk about the buzzword "human in the loop." Imagine you had a colleague that could support — how would you bring that colleague in to prevent these issues in the future, while using their time efficiently and effectively? This is a point of discussion, and the facilitator can build on their ideas and see what they're thinking.

### Part 5 — Diligence and accountability

**Alice:** Finally, step five — and I think this is really important — is judging somebody's diligence and accountability. We say: you've created this system of agents, you're going on holiday tomorrow. Would you feel comfortable launching it in its current state? And if not, how would you put systems and the right testing in place to feel confident before launching it?

**Dan:** We don't want people who are going to drive full speed into a wall with their car because they haven't learned about its capability. I love the analogy because you can definitely go faster in a car, but there's also such a thing as driving irresponsibly.

**Alice:** Exactly.

**Dan:** And not putting the right safety checks and seat belts on.

**Alice:** Exactly. So Dan, I haven't tested you.

**Dan:** I'm confident I would pass.

**Alice:** Just to share a bit about how we developed this — we started by using it with internal Brainlabbers and getting it into a state we're really happy with, and now we're rolling it out to all of our roles. Very happy to share more details on how we've used it.

## Summary

**Dan:** Excellent, and a brilliant first — as you say, I don't think there's a framework for AI hiring out there that you can read in a textbook, and we're going to have to experiment with this multiple times.

If you're hiring right now: message number one was train people up, because the people do not exist right now in this market — it's not possible. If you are hiring new people, give them a real task with AI tools and watch what they do — not whether they get the right answer, but whether they can follow the right process: whether they iterate, check the output, try a second approach, put guardrails in place. This will tell you more in 20 minutes than a CV will tell you in 20 pages.

Real enterprise AI, real company, not too much theory. See you tomorrow, unless I run out of tokens. Dan Gilbert.