Tim Barker: AI, Worthless SaaS Playbooks and Redundant Org Charts 

An actionable talk for founders who want AI to work for them.

A successful serial software entrepreneur, Tim seemed destined for a career scaling private equity backed businesses but the AI wave felt exhilarating, existential, compelling… He kept asking himself, “What happens if you’re AI‑native from day zero?”

In this talk, he shows hands-on how he runs product, marketing, and operations with a handful of people and a fleet of AI agents: the folder structures and ‘knowledge assets’ that make agents useful; how he writes specs so AI does real work instead of expensive noise; and the ‘guardian’ agents that stop automation going rogue (including the $2,000 LinkedIn mistake that taught him why they matter).

Tim’s uncovered some uncomfortable truths about how experienced software and SaaS founders must now rethink org design, funding, and their own role in an AI‑accelerated world. Tim explores what this means for you: how to think about org design, funding, and business models in a world where AI can 10x your velocity and competitors can build your features in a week. 

Beware, your hardest-won leadership skills might be the next legacy constraint you have to disrupt.

Slides

Find out more about BoS

Get details about our next conference, subscribe to our newsletter, and watch more of the great BoS Talks you hear so much about.

Transcript

The speaker before me spoke for an hour very eloquently with one slide. I’m not like that. I’ve got lots of slides, and I’m going to try and keep the energy up and the pace up – because today’s session, despite the negativity you might be feeling from the title, is going to be one of positivity. I’m here to excite you, entertain you, educate you.

This is the story of the last seven months of my life, when I went from running a 600-person business as CEO in healthcare to a 5-person AI-native business. I thought I’d share what I’ve learned along that journey – and the mistakes I’ve made. And look, I wrote this presentation a week ago. AI’s moved so quickly that half of this may already be out of date, but my job here is nothing more than a practitioner trying to figure this out as a leader.

The gift I’ve got for you at the end of this is that everything I talk about, I’ve zipped up and put on a shared drive. So you’ll be leaving with a little bootstrap guide you can take and apply for yourself if you found any of this interesting.

From 600 People to Five

The genesis for doing this was a panel session I attended in Los Angeles a few weeks ago, where one of the individuals on stage runs a one-person business doing a billion dollars in revenue. I learned nothing – other than that person was cool and doing a billion dollars.

So I’m going to give you the opposite of that. I’m going to give you something that’s admirable and actionable. I’m not yet doing a billion dollars, and I’ve got five people in the business – but the main thing was I wanted to make sure I could leave you with real, tangible things that you could see and do. When you spend time online, there are a lot of people talking about AI, but how you implement it in an organisation is a bit hard to find. You’ve got so many people giving you magic prompts, and that’s really not the answer.

So who am I? I’ve had the benefit of being in the industry long enough to have done every job apart from finance and HR. I started as a software engineer, spent probably 15 years in product, joined one of the UK’s first SaaS businesses before it was called SaaS, and ended up founding a company that became the first international acquisition Salesforce.com made. I relocated to San Francisco, became CMO for Salesforce in Europe, and then spent 10 years in startups and scale-ups.

My last business was a digital healthcare company. I took it from around 7x growth up to $80 million in revenue, IPO’d it on the London Stock Exchange, and grew from 100 to 600 people. As a CEO at that scale, you manage through other people – your job is to meet people, review plans, keep everyone aligned. You’re the pacemaker of your organisation.

After exiting, I had three choices: join another private equity-owned asset and scale it up; add some board advisory positions; or start something new. I didn’t even consider the startup route until one of the firms I was planning to advise showed me what they’d built. And over a period of about two or three weeks, I figured out – as I’m sure you have – that there is a huge market pull for everything in AI. If you’re going to learn something in AI, you have to be all in and do it right now.

So I decided to do that. My decision axis was: I want something both fun and exceptionally hard. There’s nothing better than a startup for that.

Why It’s a System, Not a Prompt

I’ve got eight things to talk about, and actionability is a key theme throughout. We are only five people, but I believe we’re doing the equivalent work of a 25–35 person company. And I’m very mindful of how I would have operated a 100-person company, so I’m building systems that can scale as the organisation does.

I’m going to talk about operationalising AI – not about having a magic prompt that makes everything 10 times better. I’m going to show you the system I’ve built, and I’ll leave you with that system to make your own. Everything I’ve learned on this, I’ve adopted from AI-first engineers and moved it into other parts of the business.

The most frustrating thing you’ll see on social media is the posts about “the 30 prompts that will change the way you work” – and behind every one of those is a subscribe button. Someone’s making serious money on Substack these days. (I do have a free Substack where you can see what I’m reading, if nothing else.)

Here’s what I’ve learned, and I couldn’t even tell you where I learned it from – except that you’re already doing it. You’re already doing it in all of your businesses. You’ve already built the foundations for an AI-first business, because AI is like the smartest person you ever hire that wanders into your business with no memory of what happened yesterday. You might remember the film Memento – a memoryless individual. Well, if you were to onboard that genius hire, you’d do the same things you do for every hire, just at speed.

You’d onboard them – give them your strategy, your values, your brand, the way you show up as an organisation. You’d give them a role. You’d make it explicit about what their role is, what their boundaries are, what autonomy they have, what authority they have, and where they need to defer decisions to others. You’d give them the work. You’d review their output, capture their progress, and then set up feedback loops to continually improve. You do all of this today.

The great thing about building an AI-first system is you just need to institute all of these things in a way that AI can use – but perhaps in a more granular way, because job descriptions especially are often far too broad. You’ve got to make them crisp and bounded.

Building Your Organisational System

In terms of the new venture I’m in – Attain IP – we are a startup focused on the legal tech industry, specifically patents and intellectual property. I joined in September full time, and the first thing we did was build our organisational system.

The irony of this is that for something as high-tech as AI, the way you do it is with your folder structure. You build five or six different folders and cascade all of your context into them.

It starts at the strategy level. You’ve already got all the assets around your organisation somewhere – you might need to reformat them into markdown files, but this is where you define your truths: strategy, brand, products, priorities for the year, the markets you serve, the markets you don’t serve, your values, and your RFCs (Requests for Comments – that’s just terminology engineers use for these documents). In the bootstrap takeaway I’m sharing today, there’s a guide that you can literally point at your public website. It’ll build a skeleton of these, and then you can fill it out with detail.

Once you’ve done that, you get your job specs done, and then your agents. There are two types of agents I have: agents that do work, and guardians that check the work. Why? Because you don’t want an agent pursuing growth at any cost – you’ve got to balance it against the trust you’re trying to build, and the economics of how you grow a business. So build guardians that check the work of agents.

You then define your missions – time-bounded projects that let you focus on getting specific things done. Those missions create outcomes and artefacts, so you need a place for those to live. The artefacts from one mission might be the inputs to another.

As you all know, context rot is real. Token windows are only so large, so you need to make sure you are always compounding everything back into assets that AI can read. That’s where your state lives – a view of where you are today, the decisions you’ve made, the work to be done, what you’ve completed.

This has allowed me to take everything I’ve learned from operating a 600-person business at scale – where, as you know, all your challenges as a CEO become people problems – and codify it into an organisation that can scale people and AI together, because your people need all these things as well.

The first thing I’ve learned in the last seven months: it’s not the magic prompt. It’s systems engineering.

The Constitution: Hard Constraints and the $2,000 Lesson

Every system has a bootstrap – a constitution. It’s the first file AI reads each session. That markdown file defines where AI should get its context, and critically, it sets your hard constraints.

One key reason you slice all your assets into different files – different missions, different agents – rather than having one super-brain, is token management. If you load everything up and have one super-brain trying to do everything, it’s going to be ten times more expensive and less effective. Sharper, specific missions and agents are the answer.

I learned why hard constraints matter the hard way. I didn’t have a constraint about AI’s permission to spend money – and it cost me $2,000.

I was rebuilding our website and thought I’d set up some skeleton LinkedIn ads to optimise later. I got an AI computer-use tool to do the setup. It went in, set everything up – just stubs, I wanted just stubs – but the thing was launching to a live website, not a staging environment. I didn’t really pay attention for two or three weeks, then looked at my credit card and realised it had switched the campaigns on. I burned through $2,000.

(I should say: I got a pretty good click-through rate on a terrible ad. So there’s that.)

The lesson: be explicit about constraints in all things – just like job descriptions for agents. Don’t soften this stuff. Make it super crisp, because that’s what makes the whole system more efficient and effective. Put a financial guardian in there. On day one.

Feedback Loops: Building a Learning System

Think about how you sit down with your team, review their work, and provide continuous feedback. That’s a system you need to industrialise at scale across all of your AI work.

2025 was the year of co-pilots, if you listened to VCs. 2026 is the year of autopilots. But as I’ll talk about more later – move at the speed of trust. I’m not quite at the point where I want agents interacting with prospects and customers directly, until I’ve got their tone of voice to a point where I’m happy with it. The ability to institute feedback loops across all these cycles is critical, and to make it a learning system over time.

Here’s an example of how we do this. We’ve got a number of agents that write in either my voice or a co-founder’s voice, so we can turn outputs into something that feels authentically us. I’ve got a markdown file that captures every asset produced – the AI-produced draft and the version I approved – and continually evaluates the delta between those and feeds that back in.

There is a compounding effect. If you do this on all the assets you produce or draft over time, your voice becomes a system asset and allows you to ultimately move from co-pilot to autopilot on some things. We’re not quite there yet, but we’ve instituted the feedback loop.

Five Examples from Practice

With all those foundations in place – the system, the pyramid of folders from strategy through to artefacts – let me give you five examples. I’m probably using 15–20 different things, but these give a good variety of how we use AI, from the minutiae through to more impactful work.

1. The Agent Operating System

I’ve got 14 agents operating today. They all have clear boundaries and governing documents. They know what they need to read when they boot up. They know how they interact with other agents. They know their boundaries, and they know when to bring in a guardian to review their work. I’ve also built some convenience commands – for example, just as you might do a stand-up at the start of the day and a wrap-up at the end, I do that with my agents too. A stand-up at the start of the day to identify what we need to do based on yesterday’s work, and a wrap-up that does a session review and writes it back to maintain a record.

I also have a project coordinator agent that looks across all the agents and wraps everything up, so I can get an overview of progress – and most importantly, see blockers and critical paths that need addressing. We’ve got over 20 missions running.

One thing I’d highlight: look at your strategy documents and realise they’re probably stale. You’ve got a good opportunity to refresh them before you start feeding them to AI.

2. LinkedIn Signal Monitor

We’re a new organisation – myself and two co-founders trying to establish our presence on the same social networks as all of you. One of our co-founders is a patent attorney of 30 years with a great network and real respect in the field, but not particularly active on LinkedIn. My job is to feed her great things to engage with.

For 20 cents a day, I’ve got an industry analyst working for me. Every morning at 8am, it goes out, looks at public conversations on LinkedIn, identifies which ones are in our domain, scores them, weeds out competitors and off-topic posts, and comes up with 10 posts she should engage with – along with recommended talking points based on her voice. We are not auto-posting. She goes in and engages directly, but the AI gives her the starting points.

This works fantastically well and is running on autopilot now. (Shoutout to Apify, which gives us API access to public LinkedIn content – that’s what makes it work at that cost.)

The feedback loop: our agent has its own email address, so it sends a daily email. All she needs to do is reply back saying “1, 3 and 5 were on point – 6 and 7 I didn’t like because of x, y, z.” When it wakes up tomorrow, it reconciles that feedback. It’s continually evolving and learning.

3. Guardians

I mentioned guardians earlier. I’ve got four guardians – essentially authorities that we run content through. These are the same things you’d have in a people-powered business: you want a design team to review marketing content, or a legal team to review a product decision. We can institute those guidelines as guardians so we don’t have to coordinate cross-functionally quite so hard.

My four guardians are:

• A messaging authority, built on April Dunford’s positioning frameworks. Everything we produce gets scored against whether it creates clear white space versus competitive noise.
• A trust evaluator. Everything we do as an organisation is ultimately about building trust. There’s a formula for trust – you can apply it and score any asset on whether it’s compounding towards trust or eroding it. We use this primarily on outbound assets.
• An ISMS guardian. As an organisation focused on intellectual property and patents, we recently achieved ISO 27001 accreditation. We have a guardian that looks across both code and marketing assets to ensure they’re accurate and aligned to those certifications.
• A financial guardian. See: the $2,000 lesson above.

You can plug guardians into any production pipeline so they automatically run and grade – and either block or flag things that need human review. Put these in place on day one.

4. Meeting Processing and the Decision Log

Like you, I meet my leadership team every week. Like you, we have an agenda. And like you, I was not as diligent as I should have been about making sure all the actions and minutes were written up.

Every call is recorded on Zoom. A few hours after the meeting, when the recording is ready, a Claude agent picks it up, processes it, and gives me a list of items to review. I determine whether we want to cascade them throughout the organisation – because some decisions you make at the leadership level need to be cascaded not just to people, but to AI systems as well. There’s a decision log that gets updated: a timeline of every decision we’ve made as a leadership team. That becomes a feedback loop into the AI models – when they restart tomorrow, they load in their latest context, whether it’s about pricing, hiring, whatever it may be.

Things you’d do today – but you have to be a bit more systemised about it, because you cannot bump into AI in the corridor and let it know about a change. You need to institute it and put it into your pyramid of context.

5. Growth Marketing Agent

Obviously we’re using AI to build our product, and I’m also using it to drive marketing. As a former CMO, I know what a good plan looks like – but I confess it’s been a few years since I had my hands on the keyboard creating LinkedIn ad campaigns.

If you look at marketing today, it seems to be two things: trying to growth-hack your way around algorithms (email spam filters, LinkedIn, advertising, SEO), and making sense of the abundance of data you’ve got – from top-of-funnel click-throughs all the way through the journey. Even for a small organisation, I already had an overwhelming amount of information.

So I have a growth marketing agent that collects all of this data every day, evaluates it alongside the experiments I’m running, and comes up with an analysis and recommendations. The biggest benefit compared to my past lives: this is completely ego-free. No one’s in there with a bias towards “I really want this campaign to work, so I’m going to highlight the good and bury the bad.” I get an objective, analytical view of my marketing performance daily, and recommendations on what to change.

I’ve got a prompt set up so I can dive into Claude to work on implementing any one of those recommendations. And the benefit is that it understands everything about my organisation – from the strategy through to the growth tactics through to the creative it helped co-produce – so it has the full context to come up with genuinely impactful analysis.

Why Old SaaS Playbooks Are Breaking

I built my career on SaaS playbooks. In fact, one reason I didn’t go down the advisory route was honestly that I thought in two or three years, my SaaS currency would run to zero. I had to reinvent myself for this next chapter.

Let me touch on a few things that are distinctly different now.

The New Law of Physics

We’ve all operated under the law of physics that software engineering headcount is the constraining function. That’s why you raise capital – the first thing you’re doing is hiring. The old model: raise, build your team, build your bench, build your revenue, raise headcount as a measure of your velocity. The first thing you probably always do when looking at a new company on LinkedIn is check how many people they have. It gives you a good proxy for how successful they are.

Not anymore.

What I’m seeing now is that AI-native organisations are high in T-shaped leaders – good breadth across what needs to be done, but not necessarily deep in one specific vertical, with the ability to instruct agents to handle execution. AI is probably going to do 60–70% of the work. It requires supervision. In fact, what it means is that every employee becomes a line manager – a line manager of AI agents. You’re instructing them, evaluating their plan, reviewing their output, and giving them feedback. All of your employees will become line managers to AI.

The unintended consequence of that: at the seed stage of funding, seed now looks like what Series B used to look like about five years ago. Because the cost to produce product was so high previously, you used to get a seed round on a great vision, a great team, and maybe a prototype. Now, seed stage means “show me the hockey stick.” The bar is continually being reset based on the fact that the cost of production is now collapsing.

From Functional Teams to Systems of Ownership

I’ve spent my career using the same hiring model: hire a great bench of C-level functional leaders, they look after specialists in their area, build the organisation out functionally. And we know the challenges that come with that – every team working with its own priorities, prioritisation friction when you need to move across teams, management overhead, alignment tax, and the need to build a large bench before anything gets done.

We’re taking a different approach. I don’t believe we’ll be operating in the same functional silos in the future. We are moving from a functional model to a systems-of-ownership model – organising so that every team can build what they need, own an outcome, and operate without being blocked by cross-functional dependencies.

I’ve got four teams – you might think of them as squads. My CTO owns the core reasoning engine, with one measurable outcome: speed, cost, and accuracy versus manual work. I own distribution – the measurable flow from someone arriving at our product to getting them activated. We’ve blurred the line between marketing website and product; people land on the site and it’s focused on converting them into usage of the reasoning engine. We have a service experience team that owns the outcome when someone is interacting with the product. And we have a sales function focused on building trust within the organisations we sell to – measured by relationship-sourced product usage.

This also helps you spot where you’ve fudged it in the past. Customer success, for example, as a standalone function is a byproduct of the SaaS model – you build an organisation to land subscribers and then stop them churning a year later. In a product-led growth approach, the logic is: give value first, ask for revenue second. Convert product users into product payers rather than putting up a paywall before they do anything. That means customer success now lives inside product. Product should own the success of its own product, and it should have commercial ownership of that.

Where the Moats Are Now

Historically, your moat might have been execution speed or the size of your sales force. But in an environment where someone can AI-develop a comparable feature in a matter of days, where does long-term sustainable defensibility come from?

Three things seem to come up consistently:

Network effects – if you’re plugged into a network, you become irreplaceable. It’s why the worst product I use every day, LinkedIn, is here to stay.

Deep domain reasoning – vertical domain expertise has huge defensive value. The depth of what you know about a specific industry or problem set is hard to replicate.

Outcomes-as-a-service – there’s been a significant shift towards services organisations using hybrid humans-in-the-loop plus AI to offer outcomes, rather than software. VCs are making big bets there.

And I’d add a fourth, given what we do: intellectual property and patents. Open AI went from zero patents three years ago to over 110, fast-tracking each to approval in ten months. Whether they’re playing offence or defence, there is a renaissance in IP to underpin innovation. If you’re building something genuinely new, it’s worth paying attention to.

New Habits for an AI-Native Day

Let me give you a view of some reflections and new habits I’ve formed adapting to an AI-native organisation.

Embrace extreme multitasking. My day starts with five shell terminals open. I give instruction to one, then spend the day jumping between them as each finishes work or hits a blocker needing my approval, working through a backlog of sessions. About half the room just nodded. At some point you realise you want to work in the terminal rather than the app because it’s just faster. It allows extreme multitasking as your work day. It’s also exhausting – because there’s nothing holding you back except yourself. Normally when you need to get work done through colleagues, at some point their eyes start rolling and the backlog wins. With AI, no one holds you back. You have to have self-constraint.

Always go into plan mode before implementation. Never one-shot into execution. Put a hard constraint to go into plan mode, challenge the plan, and only then build.

Get adversarial review from a different model. I have the same context set up in either Gemini or ChatGPT, and I take any plan and feed it in to ask for an adversarial review. I take that feedback back into Claude and then review a plan that’s been challenged by a different AI. This is really helpful – partly because in extreme-multitasking mode, I genuinely forget what I did two months ago.

Have an AI project manager. Build an agent that keeps both a forwards-looking plan and a backwards-looking record, and can run retrospectives and surface the critical path to your end deliverable.

Take deliberate breaks to consolidate. I learned this the hard way. I had so many things that were 80% automated but not finished. I started getting inundated with edge cases and business-as-usual tasks. Schedule time to pause and fully automate what isn’t quite finished. When you run an agent, you can run it ad hoc on the command line, schedule it to wake up on a cadence, or trigger it on an event. Think about which of those three is right for each thing you’re trying to automate.

Three Things I’m Still Learning

Move at the speed of trust. AI will produce something in my voice, but I’m not yet at the point where I’m letting it send that out. Move at the pace where you genuinely trust the output.

Build a system to track what you’ve done. With the volume of information and work, you need a way to avoid redoing things you’ve already done. Your AI PM agent helps here enormously.

Keep up without burning out. The pace of releases is relentless. I’m constantly hopping between tools to keep up with what’s new. That’s just the path right now.

How to Get Started in Your Organisation

As well as running a five-person company, I’m helping a few friends in larger organisations through their journey. And I don’t mean to offend anyone – but large organisations are often where good ideas go to die, because it becomes governance-first, and you move at glacial speed. I hope all of you have AI transformation in your top three priorities this year. If you haven’t, talk to me afterwards – you need to start the journey at an organisational level, not just in engineering.

A few suggestions:

Build something yourself. All of you as leaders cannot delegate this. A rule in business is never delegate your most important things to someone else, and never outsource them. So the most important thing for you right now is to become a practitioner. Build something for a passion project, or build something for the day job – just to learn. You’d be amazed at how quickly you can do this. I use two tools: Google Stitch (think of it as the Figma of AI – you write a spec, it comes up with incredible designs from day one), and then Claude Code to build it. What you can do in a few hours is remarkable.

Flip the traditional evaluation plan on its head. Don’t write a three-to-six month plan with cross-functional working groups and nine months of structured experiments before you identify what’s enterprise-ready. You know what that plan produces. Instead: leaders go first. CEOs become practitioners and model the behaviour. Work with each function leader to build and deploy something within 30 days. Don’t plan anything beyond that. Get yourself into learning and production mode, and within six weeks you’ll have a completely different perspective on what’s possible.

Benchmark your AI fluency. Zapier is doing a fantastic job of publishing their journey to an AI-native organisation. They’ve built an AI fluency rubric they use for interviewing across functions – you could use it for benchmarking your own team. Feed it into your AI and get it to help you identify how to apply it in your own organisation.

The Bootstrap Package

I know I’ve spent a lot of time on this – here’s the summary. I’m seven months into this journey. I’ve got a perspective on what every function in an organisation needs to do, and I’ve had the privilege of starting with a clean sheet of paper. I’ve been all in.

I’ve packaged two things up for you in the shared link I’ll leave you with:

The first is a set of skills – packaged commands you can load into Claude. There’s a readme file; you give it your company name and website, and it’ll read everything it can find and bootstrap an organisational model for you: the folder structure, the strategy documents, the agent specs. A subsequent skill will interview each of your C-suite leaders and build out their functional area in more detail, helping you identify the first agents you could implement.

The second is a free Substack where I share what I’m reading. I’m not publishing; I’m not charging subscriptions. You can just follow to see what’s useful.

We’re all new at this. Claude Code didn’t exist in May 2025. Everything is moving fast, and all I’ve got to offer you is a practitioner’s perspective. Good luck in your AI journey.

Q&A

Audience Member: Great talk. I’m curious about vendor lock-in. Does your business grind to a halt in those 60-to-90-minute windows when Anthropic’s servers are down? How do you think about that?

Tim Barker: Anthropic or ChatGPT being down is the new “the internet’s down” – nothing happens now when they’re down. We just review specs that AI produced for us. The more interesting question is token costs. What we’re seeing – and it’s unsurprising, you can follow Uber’s model – is that you win the market, then you introduce surge pricing. We’re already seeing that in AI. Claude’s tokens cost more from 1pm onwards when the US is awake. That’s okay – we’ll just schedule heavier work for the mornings. But it does mean you end up being a lot more conscious about context window costs. Which is why, both in engineering and across everything else I do, using the minimal amount of context needed to get the job done really matters. Otherwise your token cost is going to explode.

---

Mark Littlewood: What’s the next question you’re expecting, Tim?

Tim Barker: Oh, I honestly don’t know – kick off the crowd, I’ve got so many.

---

Audience Member: Great talk. Two questions. First, is there going to be a phone number your clients can call, or is it all automated? Second, you talked about raising money – do you actually need to raise if the economics have changed so dramatically?

Tim Barker: On the first – our goal is to still very much be a human organisation. There are real people on the website and wherever we engage, real people engage. We’re just 10x. I want to build a small organisation – maybe 10 engineering people doing the work of 100. That’s the dream: everyone highly leveraged.

On fundraising – I’ve done the raises, I’ve done the journeys, but the economics are so different now. We’re looking at a $2 million raise, which in old money is essentially peanuts – but we can get to profitability and beyond with that. And in terms of the size of the organisation I need to reach $100 million ARR, I’m looking at an order of magnitude below what it would have been. So possibly not is the answer.

Audience Member: But what would you spend it on? Tokens?

Tim Barker: You joke, but there is actually a new vanity metric emerging: tokens burned per engineer. We want a few more engineers, and we’re going to build capacity to help customers – the people change management of AI in law firms. More of the spend will be on helping our customers do their own transformation than building internally.

---

Audience Member: If someone rings you up about a patent issue, your engineers won’t be able to answer the question, right? Because engineers aren’t patent attorneys.

Tim Barker: Our engineers are patent attorneys. That’s the thing – AI has helped patent attorneys become engineers. They’re patent attorneys by training – STEM science, then cross-trained as lawyers. With AI, you can have domain expertise and engineering expertise in one person. That’s only possible now.

---

Audience Member (Meredith): You’re saying today’s seed round is like yesterday’s Series C, but also that 60–70% of the work will be done by AI. That’s roughly a 3x speed-up – which is no small thing – but it’s not the orders of magnitude implied by the difference between a Series C and a seed. Is the limiting factor really the trust boundary? The fact that if you let these things off the leash, they can do something very embarrassing – or worse?

Tim Barker: It’s a really good question, and I’ll just say: I’m still learning. I’ve learned the hard way about giving autopilot to things that should be co-pilot. There are some jobs – data analytics, business intelligence – that should have been done by computers from day one; they just weren’t smart enough. I’ll never hire a business intelligence head, because AI can do a much better job of surfacing and interpreting data across the organisation when it has full business context. But you’re right. If every new hire we onboard is going to be a line manager to AI, they’re going to figure out the boundaries just as any good manager would – where to give full autonomy versus where to stay close. The VC view is that everyone becomes 10x. We may not be at 10x right now, but we’re somewhere between 3x and 5x, and it’s trending upwards. The key learning is: you still need governance structures. Once you have that governance in place, over time you can apply different levels of oversight – autopilot in some areas, very hands-on in others.

---

Audience Member (Bruce): When you put up the slide saying “CXOs need to go first,” I thought you meant get rid of them first and replace them with AI. My question is: let’s say it’s a year from now and you’ve raised the $2 million. What does your org chart look like? Which functions still have humans with a staff of agents? Which ones do you combine or eliminate?

Tim Barker: Let me answer that through the lens of our first senior engineering hire, because that skill set maps across everywhere. Our first senior engineer hadn’t written a line of code in five years. He’d reached the level where he was managing a team as a senior architect – spending his days reviewing work, challenging plans, and reviewing outputs. That’s exactly who we wanted. T-shaped: they know what good looks like, they know what a good strategy looks like, they can review and challenge, but they’re not hands on the keyboard doing the execution. That’s the profile we want across every function.

Audience Member: So you’d still have heads of finance, sales, marketing?

Tim Barker: Yes, but they’d be living in a system. They might not have a traditional department under them – they own an outcome. In sales, for example, the traditional SaaS model became a commissioning model at every point of the funnel. I’m building product-led growth, so the sales function is focused on conversion: marketing surfaces a user, sales converts them into a subscriber. The job spec is different, but we’ll have the same broad bench. The difference is they’ll be in an organisational unit that has all the resources to succeed, rather than having to operate cross-functionally and fight for priority.

Audience Member: But if you’ve got someone running marketing and someone running sales – one surfacing leads, one closing them – there’s still potential for cross-functional friction.

Tim Barker: Of course there is. Which is why you need to systemise – because it’s not just people friction you’ll have now, it’s agent friction too. You have to make sure the feedback loops between those roles are systemised, and that the boundaries are really crisp.

---

Mark Littlewood: I’ve got about seven more people who want to ask questions, and I’m going to have to draw a line under it – not because I don’t want to hear them, but because we’re out of time. Tim, thank you. Wow.

---

---

Online workshops

New to AI in your business?

AI in Your Business: Where to Start

Set up your AI company OS. Two sessions, cohort of 12.

“I’ve been talking to it loads about positioning, sales framing, North Star metrics. I came up with a succinct statement that I really like.”

Marv Rance, InTouch

Register now Want to do it with just you? One-on-one option ›

Already set up your OS?

Building Your AI Company
Operating System

Not sure what to do next. Four weeks, cohort of 12, with Tim Barker and Mark Littlewood.

“I think about the business differently than I used to. I would never have come to that conclusion without this process.”

Stephen, cohort participant

Register now

Can’t make it? More dates coming up