The line between software and service is dissolving. The businesses that don’t understand what’s replacing it are going to be left holding infrastructure that nobody wants to pay for.
I spend my time sitting at the intersection of two worlds that are about to become one.
On one side: software businesses. SaaS platforms, tools, systems — companies that sell access to technology at a margin that would make a traditional service business weep. On the other side: service businesses. Agencies, consultancies, managed service providers — companies built on human expertise, delivered at a rate card, and constrained by the brutal economics of time.
For twenty years, those two worlds operated on different planets. The software people talked about ARR and NRR and churn. The service people talked about utilisation rates and day rates and project margins. They admired each other from a distance. Occasionally they competed for the same client budget. But they were fundamentally different things.
That’s over.
What I’m watching happen — from boardrooms, from strategy sessions, from the conversations I’m having with founders and investors and operators across both sectors — is something I’ve started calling The Great Convergence. The line between software and service is dissolving. And the businesses that don’t understand what’s replacing it are going to be left holding infrastructure that nobody wants to pay for.
The filing cabinet problem
To understand The Great Convergence, you have to understand what enterprise software actually sold for the last two decades.
It sold storage of truth. One authoritative place where your customer data lived, your employee records, your financial transactions. The CRM. The HRIS. The ERP. These weren’t beautiful pieces of software — anyone who’s spent meaningful time inside a Salesforce deployment knows they’re anything but. They were canonical. The VP of Sales opened the CRM not because she loved it, but because it was the place. The single source of truth that made the chaos of a growing organisation legible.
Salesforce is the clearest expression of this model. A $41 billion revenue business, 150,000 enterprise customers, 22% global CRM market share. Three times its nearest competitor. The most important software company of its generation, built almost entirely on one proposition: your customer data lives here, and leaving would be unthinkable.
The moat wasn’t the software. It was the switching cost. Years of customisation. Thousands of integrations. Tens of thousands of trained users whose daily habits were built around a particular interface. The humans were the lock-in.
That’s a brilliant business model. Until the humans start to disappear.
What Benioff actually admitted
In September 2025, Marc Benioff went on record with Fortune and said Salesforce had just cut roughly 4,000 customer service jobs. AI agents now handle 50% of all customer interactions at the company. Support costs dropped 17%.
The CEO of the world’s largest CRM company — the company that sells customer relationship software to 150,000 enterprises — publicly confirmed that AI agents are eliminating the humans who use his software.
This is not a small thing. This is the person with the most to lose from this narrative saying it out loud anyway, because the results demanded it.
Within six months, Salesforce stock fell 26%. The broader software index dropped 22%. What the market quickly labelled the SaaSpocalypse erased somewhere between one and two trillion dollars in market capitalisation from the software sector in under 60 days. Atlassian reported its first ever decline in enterprise seat count. The iShares software ETF crashed from $117 to $82.
The mainstream interpretation was that AI is killing software. That ChatGPT will replace Salesforce and Claude will replace Workday. That reading is mostly wrong. But the crash is completely real.
What’s actually happening is more interesting, and more consequential, than a replacement story.
Value doesn’t disappear. It migrates.
The value embedded in enterprise software isn’t vanishing. It’s moving — from one layer of the stack to another. From the system that stores data to the system that does work.
Salesforce’s per-user pricing runs at roughly $13,600 per user per year when you factor in add-ons and professional services. That model was always a proxy for human activity. You paid per seat because each seat represented a human being turning up to work, opening the system, logging calls, updating records, generating pipeline.
When an AI agent handles 50% of that activity, you don’t need 50% of the seats. And when you don’t need the seats, you don’t need the lock-in that came with them. Agents don’t have favourite dashboards. They don’t have trained habits. They call APIs. They can call a different API tomorrow.
The humans were the moat. Remove the humans, and the moat empties.
This is where Salesforce’s structural problem becomes visible. Its entire business was optimised around being between the human and the work. A filing cabinet for the information those humans needed to do their jobs. In a world where agents do the job directly, being the filing cabinet is the least defensible position in the chain.
Salesforce is worth $180 billion today. It peaked at $350 billion in December 2024. Roughly $170 billion in market capitalisation — gone in fifteen months. Revenue growth has collapsed from 25% annually to under 9%. Margins have expanded, but only because headcount fell. That’s not a growth story. That’s a harvest.
The convergence I’m actually watching
Here’s where I need to bring this back to what I’m seeing in the real world, because the Salesforce story is the high-profile proof point of something happening at every level of the market.
I work across a mix of software businesses and service businesses. And the conversation I’m having in both, with increasing frequency, is the same conversation.
In software businesses, the question is: if our users are being replaced by agents, what do we actually sell? The per-seat model is under pressure. The UX layer — the interface humans used to interact with the system — is being commoditised by AI that can sit in front of any system. The data layer is increasingly being pulled by agents rather than navigated by humans. If your value proposition was “we make it easy for people to do X,” and people are no longer doing X, you have a problem.
In service businesses, the question is almost the inverse: if we can now deliver outcomes without proportional headcount growth, who captures that margin? A digital agency that used to need twenty people to run an SEO function can increasingly do it with five people and a stack of AI tooling. That’s either a disaster for the agency model or a fundamental repricing of what expertise is worth — depending entirely on whether the agency understands what it’s actually selling.
What both types of business are converging on, whether they know it or not, is the same thing: systems of work.
Not software. Not service. Work. The actual execution of business outcomes — qualified leads, resolved support tickets, closed renewals, filed accounts, approved budgets. The thing that happens, not the thing that records it.
Who owns the work owns the moat
The new moat in enterprise technology is ownership of the workflow. Not the database. The process.
ServiceNow is the clearest example of a company that understood this before the rest of the market did. It was never primarily a system of record. It was always a workflow engine — routing, automating, escalating, tracking. Built to orchestrate work across departments. In an AI-native world, that starting position is extraordinarily valuable, because agents need orchestration, and ServiceNow is already in the orchestration business.
ServiceNow powers 80 billion workflows annually for 85% of the Fortune 500. That installed base of workflow definitions, routing logic, and orchestration patterns is the new form of lock-in. It’s far more defensible in an AI-native world than a database of customer records, because it’s not just where the data lives — it’s how the organisation thinks.
And this is the piece I find most interesting, and most underappreciated.
Every time an AI agent executes a workflow, it generates something I’d call a decision trace. Not just the outcome — the context, the reasoning, the conditions under which a particular choice was made. Why the discount was approved. Why the lead was disqualified. Why the escalation happened at that point and not earlier. That decision trace, accumulated over thousands of executions, becomes structured, queryable organisational intelligence.
The next renewal, the next lead qualification, the next support escalation — the agent draws on every prior decision trace to handle it better. The flywheel is not just that AI gets smarter. It’s that the organisation’s way of working becomes encoded, and that encoding is owned by whoever was in the execution path when the decisions were made.
You cannot reconstruct decision context after the fact. You have to be present when the decision happens. This is why owning the workflow is so valuable. The old moat was we have your data and leaving is painful. The new moat will be we have learned how your organisation actually works, and no one else has.
What this means for businesses right now
I want to be direct about what I think this means in practice, because The Great Convergence isn’t a future scenario. It’s happening in the businesses I sit across right now.
For software businesses: the question is no longer how many seats you can sell. It’s whether you’re in the execution path. Are you the system that records what happened, or the system that makes it happen? The former is becoming a commoditised data layer. The latter is where the value is accruing. Agentforce — Salesforce’s attempt to get into this game — has reached $800 million in AI-related ARR. That’s real traction. But consumption-based pricing on AI interactions produces lower and less predictable revenue than seat licensing, and whether it can replace what’s being lost remains entirely unproven.
For service businesses: the question is whether you’re pricing your expertise or your hours. If you’re still pricing hours, you’re selling the thing AI will commoditise fastest. If you can price outcomes — the closed deal, the recovered revenue, the qualified pipeline — then you’re positioned for a world where the labour input is increasingly invisible and the result is what gets purchased. The total addressable market for AI-driven service delivery isn’t the software market. It’s the labour market. And the labour market is six times larger.
For everyone: the convergence means that the categories we’ve used to organise the market — software company, service company, technology business, professional services — are going to be less useful than they were. The businesses that win the next decade will be the ones that own outcomes. The ones that can say: we don’t just store the record of what happened. We made it happen. We’re making the next one happen right now.
Salesforce will not go to zero. Companies with $41 billion in recurring revenue do not vanish. But they do transition — from defining how enterprises work, to supporting how enterprises work. From the operating system of the sales organisation to a backend data service that AI agents query through APIs. That transition from the centre to the periphery is one of the most significant value migrations in the history of enterprise technology, and it’s happening faster than almost anyone expected.
The filing cabinet was the moat once.
We don’t need filing cabinets anymore. We need systems that empty them.
Simon Penson is the founder of Scaled Consulting and serves as a Non-Executive Director across a portfolio of agency, SaaS, and B2B service businesses. Scaled works with ambitious B2B companies to build the commercial infrastructure to scale — and exit — on their own terms. scaled.co.uk
