Five unsolved problems worth building a career on

I’ve led Talent at Attio as we’ve grown into a team of 200+. The best part of my job is meeting candidates excited about our hardest problems and employees solving them. As agents do more of the work, systems of record become more valuable. That transition is creating some of the hardest problems in software, and extraordinary career opportunities for the people who solve them.

I spend my days sharing our biggest challenges with exceptional people across Europe, the US and now APAC. But things are moving so quickly that I wanted to share five of the most interesting problems we're working on at Attio, and why they're worth building a career on.

For context, there's an argument that the system of record is the old idea: a productivity tool people logged into to track their work. But the opposite is true. The system of record is the foundation agents run on. It holds the context they read, the guardrails they work within, the tools they act through and the feedback that tells them whether it worked.

The models will keep changing. The record is the part that remembers.

Building software was already difficult when you had two types of users: builders who wanted complete flexibility and people who wanted simplicity. We call them dark and light users. Now there's a third: agents that need to read, reason and act across the same data, often without human intervention. And also a fourth dimension: your customer's customers, who interact directly with the agents built to serve them.

You cannot build the best product by picking a side. Collapse toward the dark builder and you've added complexity for a light user; collapse toward simplicity, and the builder leaves for something they can actually shape to their unique use case.

As agents become more autonomous, the value shifts from the interface where people work to the infrastructure that determines what agents know, what they're allowed to do and whether their actions can be trusted.

Agents aren't a feature bolted on, they're a new layer of users working internally and externally. We serve billions of API calls, millions of MCP tool actions and tens of thousands of custom agents.

Attio is moving from building tools for the team that serves customers to building agents that prospect, support and save accounts directly for them. Every one of those interactions is only as good as the record behind it.

We're figuring this out while our product and business expand to serve different types and sizes of companies, globally. That opens us up to even greater complexity, technical requirements, use cases, usage and, of course, expectations.

It blows my mind in the same way that trying to comprehend the size of the universe does.

Five problems worth building a career on

1. Engineering | Permissions and the write path

How do you let thousands of agents act autonomously without compromising trust?

Human error has always existed in software. What's changing is the speed, scale and autonomy with which mistakes can propagate. An agent can write incorrect values across thousands of records before a human even knows something has gone wrong.

Permissions are a clear example of why it's really hard to build Attio. It has to be solved across every object, every surface, and every actor at once. Whether your team is ten humans or ten thousand agents, it's the same problem: nothing should act on context it shouldn't see. That's why row-level permissions are one of the biggest things on our roadmap. Then consider what holds it together when agents run unsupervised. What about conflict resolution when two agents and a person touch a record in the same second?

Solving these problems is changing how we build software in the first place. We're moving towards engineers spending less time writing and reviewing code and more time designing the architecture and guardrails, without lowering the craft required and quality bar.

Figuring all of this out is something our engineering team are running towards.

2. Design | Distinction, now with a mystery user

How do you make complex agent behaviour understandable without making the product complex?

Looking at a record, what a person said out loud on a call and what a model inferred from context needs distinction and an experience that builds trust.

Every record UI assumes a human made every edit, because until recently they did. Now it carries another meaning (who wrote this, how confident is it, has anyone checked), and that leads to different experiences: some invisible, others optimised for the human task, all constructed by the builder of the workflow. When an agent proposes 400 changes, what's the surface for approving them?

The goal isn't maximum autonomy. Each customer needs to choose how much autonomy to give at each step, and the design has to make that choice feel obvious. Some of these experiences won't have a screen at all.

There’s no dominant design pattern for making outcomes and progress feel inevitable, yet.

3. Commercial | Pricing the product

What should software cost when the amount of work no longer correlates with the number of people?

Per-seat pricing is a bet that humans are the unit of value. That bet doesn't hold on its own anymore. Do we need a platform fee for the shared context Attio provides? Should we keep seats for the humans? How many credits do you need for the work agents do?

Outcome pricing sounds better until you try to define an outcome across a whole world of use cases, business sizes and revenue models. Credits have to be flexible enough to keep processes running through a surge, with controls over who spends them and on what. And how do you price an agent that talks directly to your customer's customers?

Companies that get this right will define how software is monetised for the next decade.

4. GTM | The sales motion

How do you sell the value of a product that customers need to build with before they fully understand it?

For years, selling a CRM was a "why us" conversation. Everyone knew what a CRM was. Now it's a "why one" conversation for Attio, because we're explaining a category that didn't exist a year ago. Customers come to us knowing AI should change their go-to-market and we can show them how.

When a purchase decision is made, many others are supporting, watching and awaiting the decision. The buyer of a system of record is buying for the builders, the users and the agents. How do you demo what's possible without the prompts or the data in place? One of our reps answered that for a customer recently, demoing a fully populated instance, working web pages and a script timed to the minute. It advanced the buying decision by six months. Another example is our Solutions and Sales teams now building SDK apps live on calls with customers.

Your sales ability needs to be as flexible and technical as the product.

5. Talent | Hiring the right people

This is the one I live daily.

How do you hire exceptional people for problems nobody has solved before?

Our world is moving so fast that every role comes with some speculation, which makes calibration and commitment hard on both sides. But, quite wonderfully, it results in mission-driven, ambitious people who have great judgement.

Screening the profile doesn't add enough signal. But having great conversations is how you set the bar. The strongest signal I find, repeatedly, is people who have built systems. Often from a place of frustration that things should just be better. This drive helps me predict someone's performance in ambiguous territory better than brand signals or credentials.

I look for three things beyond the skills required for the role.

• Drive: people who create change rather than wait to be asked.

• Judgement: people who can make good decisions with incomplete information.

• Systems thinking: people who improve the way work happens, rather than just complete tasks.

The flip side is the opportunity. Very few people will get the chance to solve problems like these from the inside, at the moment they're emerging. The people who do won't just build valuable skills. They'll learn to become force multipliers; people capable of transforming how entire organisations operate. And I believe they'll be the ones leading the next generation of companies.

It's the combination of drive, judgement and skill that makes people invaluable when the problems haven't been solved before.

Join us

The next generation of great software companies won't just provide tools for people to do their work. They'll provide the context, permissions and infrastructure for humans and agents to work together.

The Engineering, GTM, Marketing, Operations and Product challenges this creates are largely unsolved. That's what makes this such an extraordinary time to build at Attio. We're not just hiring people to scale what already works. We're hiring people to figure out what comes next.

attio.com/careers