The Agentic Stack: A UK Founder's Map of the £100bn Opportunity Hiding Inside Legacy SaaS

Most market maps are drawn by analysts. This one is drawn by a founder who is shipping into seven of the categories on it. The bias is operational — I care about which category replaces which incumbent, on what timeline, with what unit economics, and where the UK founder window is widest. The bias is also useful, because it forces the framing into "what would you actually build?" rather than "what would you put on a slide?"

The headline number — £100bn of UK SaaS spend that is structurally vulnerable to agentic replacement — is an aggregate. It is not the total UK SaaS market (which is larger). It is the portion of UK software spend that sits in the workflow layer: tools that ask a human to operate them, on top of a system of record that survives.

1. The two-layer model: systems of record vs workflow tools

Modern SaaS stacks have two layers. Underneath is the system of record — Salesforce holds the customer, Xero holds the ledger, Workday holds the employee, Shopify holds the catalogue. These layers are durable. They do not get eaten by agents in this cycle because they are infrastructure, not workflow.

On top of the system of record is the workflow layer — Klaviyo operates email on top of Shopify, Chaser operates credit control on top of Xero, Zendesk operates support on top of Salesforce. This is the layer that asks humans to do the work. This is the layer that is replaced.

Across the UK SME and mid-market segment, the workflow layer is roughly 65–75% of total SaaS spend. That is the £100bn we're mapping.

2. The eight categories, ranked by agentic readiness × vulnerability

1. Lifecycle email & SMS (Klaviyo, Mailchimp, HubSpot Email). Highest readiness. Outcome (revenue produced) is hard, attribution is solved, the operator hours are enormous. Replaced by agentic email — see AutoEmails.
2. Credit control (Chaser, Satago, manual chase spreadsheets). Equally high readiness. £20bn of UK SME late payment is the prize. Replaced by agentic finance — see AutoSettle.
3. Competitive intelligence (Crayon, Klue, SimilarWeb). The lowest-satisfaction category in marketing, structurally agentic. Replaced — see Competable.
4. Ecommerce checkout (Shopify default checkout + upsell apps). The highest-leverage square inches of any DTC business. Replaced by agentic checkout — see AutoTill.
5. Helpdesk & ticket triage (Zendesk, Intercom, Freshdesk). Bounded judgement, repetitive, complete written record. The category leaders are now scrambling to ship agent-first products. The challenger window narrows fastest here.
6. Ad operations (Google Ads, Meta Ads management, agency seat layers). Performance Max already does much of the work — the agentic version sits above the platforms and owns budget allocation, creative iteration, and reporting end-to-end.
7. HR ops (BambooHR, Personio, manual processes).Onboarding, compliance, references, leave management. Slower-moving market but textbook agentic shape.
8. CRM operator workflows (the layer above Salesforce/HubSpot CRM). Sales-engagement tools (Outreach, Salesloft) are the kill zone — the system of record (CRM itself) survives.

3. The pattern that repeats across all eight

Each category passes the same five tests:

• Rules-based escalation logic that humans currently operate manually.
• Repetitive written or transactional comms that LLMs handle at scale.
• Complete written record (no missing context for the agent).
• Hard, measurable outcome (cash collected, revenue produced, ticket resolved).
• Bounded judgement — strategy stays human, tactics go to the agent.

And each category currently sits at low buyer satisfaction with the incumbent — operator fatigue, dashboard fatigue, retainer fatigue. That is the door agents walk through.

4. Why the incumbents can't ship the replacement themselves

The repeated question I get is "won't Klaviyo / Crayon / Chaser / Zendesk just ship this?" The answer is no, structurally, because every public company's AI roadmap is constrained by one rule: do not cannibalise the seat-licence revenue you are publicly reporting. Every incumbent ships AI as a feature inside the existing pricing model. None of them ship an agent that replaces the operator role their seat economics depend on.

That is the founder window. It does not stay open forever — by 2028 either the incumbents bite the bullet or new public challengers (us, others) take the categories. But for 2026–2028 it is the largest founder arbitrage in software.

5. The UK as the right launch market

The agentic-replacement playbook works best in markets that: buy on payback period (not feature richness); have high SME density (so the deal cycle is short); have a finance culture that respects outcome-based contracts; and have a customer- acquisition cost lower than the US.

The UK ticks all four. SME density is among the highest in the developed world. The finance culture has always been outcome- anchored. Distribution costs in the UK SME segment are roughly 40% of the equivalent US cost. This is why Autothink is being built UK-first — not patriotism, just unit economics.

6. What you should do, depending on who you are

If you are a founder — pick one category from the eight, build the agent that owns the outcome end-to-end, price on outcome, distribute through whatever existing customer relationships you have. Distribution is the moat. The technology is not.

If you are an operator — audit your stack and flag the workflow tools sitting on top of systems of record. Each one is a candidate for agentic replacement before 2028. Plan the transition before it plans you.

If you are an investor — the agentic stack is the largest software opportunity since cloud. The bet you want is on the founders building category-defining outcomes in the eight categories above. Avoid AI features bolted onto existing SaaS — that is the bet the public markets are about to re-price down.

To talk through any of this, book a call. To go deeper on the operating philosophy behind Autothink, read Systems, Not Features.