You Do Not Need an AI Automation Agency (or an Agent Ops Hire). Here's What to Do Instead.

On 20 April 2026, Aaron Levie, co-founder and CEO of Box, posted a thread that went round the operator internet fast. His argument was clean.

The jump from working with a chatbot to having an agent that actually automates a business process requires a real amount of work. Most companies will need dedicated people responsible for bringing automation to their teams, rather than leaving it to every individual employee. Partly because the work is more technical than people imagine, partly because it is hard to do as a side project.

The job spec, in his framing, covers mapping new workflows with agents, implementing deployment systems, wiring up internal systems, building evals, figuring out where the human is in the loop, managing upgrades, and handling the change management of the existing process. He estimated 500,000 to 1,000,000 jobs in this category will be created over the next five years. Different companies will call it different things: AI Automation Lead, Head of AI Automation, Agent Ops, AI Experience Engineer, AI Solutions Architect.

Levie is mostly right. He is writing for enterprise buyers, and for enterprise buyers, he is correct. For founder-led businesses under about £5 million in revenue, his conclusion is mostly wrong, and acting on it is expensive.

This piece is for the second group.

Why Levie is right, for large enterprises

Start with where the thread lands cleanly.

At enterprise scale, agent deployment is a specialist discipline. Large companies have sprawling tool stacks, legacy integrations, compliance requirements, approval workflows, and dozens of business functions each with their own context. Wiring an agent into that environment, feeding it the right data, building evals that catch failure modes, and managing the change process across a workforce of thousands is not a side-of-desk task. It needs a dedicated function. In a company with 2,000 employees and 200 software systems, the Head of AI Automation is not an optional hire. It is a structural necessity.

One reply to Levie's thread, from @XataAndCo Research Team, put it usefully. The people the large companies are hiring are not product managers. They are systems integrators for AI failure modes. That framing is correct, and the bottleneck is real. At that scale, you are paying someone full time to absorb the operational risk of agents going wrong in production.

The agentic market is projected to hit $52.6 billion by 2030 according to multiple analyst reports circulating around that thread, and most of the value is in deployment rather than the model itself. Enterprises that get deployment right will compound. The ones that do not will fall behind. Levie is telling enterprise leaders to staff up, and he is correct.

Why it is wrong for a £2M business

Now shift the lens.

You run a £2M business. Maybe you are a recruitment firm with twelve people. Maybe you are an accountancy with eight, a law firm with five partners, a training provider doing £1.8M in revenue. You read Levie's thread. You think, "we probably need one of these AI Automation people."

You almost certainly do not, and here is why.

The problem does not scale to a full-time role. You do not have 200 software systems. You have maybe eight. You do not have dozens of business functions; you have three or four. The automation work is real, but it is bounded. A good operator can map, deploy, and evaluate the agentic workflows in your business inside two or three months. After that, you are maintaining rather than building. A full-time hire whose job is "AI automation" ends up with too little to do, and the temptation to justify the role by adding complexity you do not need is enormous.

The talent market is thin and mispriced. Job titles like Head of AI Automation and Agent Ops are so new that search volume for them in the UK is effectively zero. There is no established career ladder, no standardised comp bands, no tested interview process. The people credibly doing this work well are either at enterprises earning £120,000-plus with equity, or they are running their own practices charging daily rates that imply annualised comp far above that. Recruiting one at a price your business can support is a slow, uncertain hunt. By the time you have hired and onboarded, you are nine months behind where you could have been.

You need the judgment, not the title. The Levie job spec breaks down into two kinds of work: the technical (deploying, wiring, evals, upgrades) and the operational (change management, workflow design, defining the human in the loop). The technical work is compressible with good tools. The operational work is the same work a fractional COO already does. If you hire for the title, you get a specialist whose technical half is only 40 per cent utilised and whose operational half is often weaker than someone who has actually run businesses.

You build the wrong shape of dependency. Hiring a dedicated AI automation lead in a £2M business puts one person in charge of the rails you will depend on for every function. If they leave, you are exposed. If they stay and build, the temptation to over-engineer is strong. The automations become brittle, specific to how that person thinks, and hard for the rest of the team to modify. You wanted leverage. You built a bottleneck.

This is the same argument as the fractional COO case. At your scale, hiring a full-time specialist for a role that does not yet have full-time weight is how founder-led businesses get operationally stuck. A fractional model is better shaped to the problem.

What it actually costs to hire an AI automation agency in the UK

Let us price the alternatives honestly.

AI automation agencies are the most-marketed option. Search volume is there; supply has exploded in the last eighteen months. UK engagements in 2026 typically look like this.

You are paying for a team rather than a specialist. The team usually includes a sales person, a project manager, a developer, and possibly a strategist. Three of the four do not touch your business directly. The overhead is real, and the day-rate maths implies you are paying roughly 1.5 to 2 times what the underlying operator costs.

The better agencies deliver. They have standardised playbooks, a tested stack, and institutional memory from having shipped dozens of similar builds. The worse ones white-label a Make.com account, paste in a Claude prompt, and hand you a Notion doc. You cannot tell the difference from the pitch deck. You can tell the difference six weeks in.

What it actually costs to hire an Agent Ops lead in-house

If you go in-house instead, the sticker price looks lower. It is not.

That is a lot of committed cost for a role whose market wage is unsettled and whose workload in a £2M business is bounded.

There is a recruitment risk on top. Referring back to the thread, one reply from @ronnieschaniel observed that upskilling existing developers often works as well as hiring new. That is true, if you have developers. Most founder-led professional services businesses do not. You are starting cold.

The honest calculation: hiring a dedicated AI automation lead is a £100,000-plus annual commitment for a role that, at your scale, has about £40,000 of actual work in it for the first eighteen months.

The fractional operator alternative

There is a third option that most of the thread missed, and it is the one that actually fits the £1M to £5M founder-led business.

Hire a fractional operator who already has the AI stack wired in.

The fractional operator is not a new role. It is an extension of the fractional COO model, which has been growing for the last three years and is now mature. The difference in 2026 is that the fractional operators worth hiring have built their practice around AI from the start. They have tested stacks. They have deployed agents across multiple businesses. They know which use cases actually land and which ones sound good in a pitch deck and die in production.

You get one hire, not two. You get the operational judgment and the technical execution in the same person. You pay a retainer of £4,000 to £15,000 per month rather than a £100,000-plus fully loaded salary, and you do not commit to capacity you will not use.

This is the argument I make in the companion piece, What a Fractional COO Actually Does (And When to Hire One). The Levie thread is useful because it forces the question. For enterprise buyers, the answer is to staff up. For founder-led businesses at your scale, the answer is that your next hire is a fractional operator, and that hire absorbs 90 per cent of the AI Automation Lead job spec as a natural part of the role.

What the fractional operator's AI stack actually looks like

To make the argument concrete, here is the working stack I use across engagements.

Thinking and writing layer. Claude for analysis, document drafting, research synthesis, SOP generation, meeting summary expansion, internal memos, client-facing reports. This is the single highest-leverage tool in the stack. If your fractional operator is not using Claude as a daily work surface, they are behind.

Capture layer. Granola for meeting transcription. Calls turn into searchable records automatically. Paired with Claude, that means coaching sessions, founder check-ins, and team meetings feed directly into written artefacts with no additional effort.

Workflow automation layer. Make.com for connecting systems. Triggering actions across tools, reducing manual steps, building the middle layer that turns "we should automate this" into shipped automation. For simpler cases, native Zapier or Pipedream; for more complex cases, custom builds.

Structured data layer. Airtable for most cases. Supabase where the business needs a proper database with row-level security and custom interfaces. This is the operational source of truth, the place reports and dashboards actually pull from.

Interface layer. Vercel and Next.js where a custom dashboard or internal tool is justified. Usually not needed in the first six months. Useful once a business has standardised its data and wants a clean operator surface.

Evals and guardrails. Custom, tailored to the workflow. Evals are not yet a product category you can buy off the shelf for SMB. This is where the fractional operator's judgment matters most: deciding when a workflow is safe to run unattended and what failure modes to watch for. This is exactly the work @XataAndCo described as "systems integrators for AI failure modes". In a lean business, that work is one layer of the fractional operator's job, not a full-time hire.

An experienced fractional operator brings this entire stack, battle-tested across multiple businesses, on day one. You do not have to assemble it, nor pay someone full time to maintain it.

When you will actually need to hire for this

None of the above argues that the Levie prediction is wrong forever. It argues that it is wrong for you, now.

Here is when the calculus flips and you should hire dedicated AI automation capacity.

You cross about £10M in revenue. Complexity growth is not linear with revenue. At £10M you typically have enough tools, workflows, and functions that keeping a dedicated person busy is realistic. Below that, it is not.

You are running 20-plus agentic workflows in production. At that volume the maintenance load alone, upgrades, regressions, prompt drift, tool changes, stops fitting inside the fractional operator's time budget. You need someone whose primary job is keeping the agents running.

You have regulatory or compliance exposure that changes the risk profile. Legal, healthcare, financial services where an agent getting it wrong has material downside. At that point the evals and guardrails work is a dedicated role, not a shared one.

Your fractional operator tells you it is time. This is the most reliable signal. A good fractional operator will flag when the work has outgrown the model, rather than stretching the engagement to protect their retainer. If yours does not have that conversation with you, they are the wrong one.

Until one of those four conditions is true, you do not need a full-time AI Automation Lead. You need a fractional operator who already is one.

FAQ

Is hiring an AI automation agency ever the right answer for a small business?
Sometimes. If you have a single well-defined build and no interest in an ongoing operational relationship, a project-based engagement with a reputable agency can work. The failure mode is signing an open-ended retainer when what you actually needed was one workflow.

What is the difference between an AI implementation consultant and a fractional operator?
Consultants advise and produce recommendations. A fractional operator builds and owns. If the deliverable you are buying is a slide deck, you have a consultant. If the deliverable is a working system the team uses daily, you have an operator.

Can Levie's prediction still be right for my business five years from now?
Quite possibly. If your business is at £2M now and at £10M in 2031, a Head of AI Automation might be exactly the right hire at that point. The argument is about sequencing, not about whether the role is real.

What happens to the fractional operator when the business crosses the threshold and hires in-house?
Good fractional engagements plan for that. The operator hands over their stack, their runbooks, and their evals, and either transitions out or moves into a true operating partner / advisor role. If your fractional operator is not building towards a handover-ready state, that is a red flag.

What about the @Ja4h3ad point in the thread, that the right title is "AI Experience Engineer" and user experience is part of the job?
Fair. Conversational design and user experience are real disciplines. In a £2M business the volume of that work is small and the fractional operator can cover it. In a larger business, UX of agents is a specialist lane and worth staffing separately.

Source thread: Aaron Levie on X, 20 April 2026. View original. Credit to @XataAndCo Research Team for the "systems integrators for AI failure modes" framing, @Ja4h3ad for the UX point, and @ronnieschaniel for the upskilling alternative.