How Business Software Survives the Agentic Era

Something unusual happened to the software industry at the start of 2026. Within about 30 days, roughly two trillion dollars in market capitalisation evaporated from companies that had spent the previous decade building what looked like unassailable positions in enterprise software. Atlassian dropped 35% after reporting that enterprise seat counts had declined for the first time in the company’s history. Salesforce fell 28% despite growing revenue. Workday lost around 22% year-to-date, compounding an already difficult year. Its founder returned to the CEO role to steady the ship.

The catalyst was not a recession, a rate shock, or a regulatory crackdown. It was agentic AI. Specifically, it was the dawning realisation that the autonomous software agents now being built and deployed at scale can do the work that justifies billions of dollars in per-seat subscription revenue, without the seats.

A single story circulated widely and captured the moment perfectly. A former Amazon executive built an entire CRM system over a single weekend. One Salesforce customer terminated a £350k contract to replace it with a custom solution they built themselves on a vibe-coding platform. Neither of these is a statistical sample. But they both point at the same structural shift, and the market is pricing it in.

The question this article wants to answer is not whether SaaS is dying. It is which parts of these businesses still have genuine value, and what the survivors need to do about it right now.

What actually got disrupted

To understand what is happening, you need to be clear about what enterprise software was actually selling.

The value proposition of a mature ERP or CRM was never just the features list. It was three things bundled together. First, a system of record, a place where your business data lived, accumulated, and became an asset. Second, business logic, the encoded rules about how your business worked, tax rules, approval chains, compliance requirements, process flows built up over years of implementation. Third, a user interface, the screens and dashboards that let your people interact with the data and the logic.

Agentic AI has attacked the third layer comprehensively and is making serious inroads into the second. The first, the data itself, remains the most defensible asset these companies own. Possibly the only one.

The screens are the first casualty. If an AI agent can query your CRM, interpret the data, surface insights, and take action, directly and in natural language, then the carefully designed dashboards that used to justify your licence fee become friction. Not value. The interaction layer has been commoditised. Any business that still thinks its competitive moat is a well-designed UI is in serious trouble.

The business logic layer is more complicated, and this is where the most interesting strategic question sits.

The paradox of the rules

There is a prevailing narrative that agentic AI is going to replace the business logic embedded in enterprise systems. In some cases that is true. But there is an equally powerful counter-argument that gets much less attention.

The most capable agentic systems being built today are not pure AI. They are hybrid systems that wrap non-deterministic AI models in deterministic infrastructure. The AI handles the parts of a workflow that require judgment, synthesis, and natural language. The hard-coded rules handle everything else. You do not want an AI to hallucinate a new tax classification. You do not want it to decide on a whim to skip a regulatory approval step. You want it to call into the compliance engine that already knows the right answer and has an audit trail.

This means the 30 years of business logic baked into a mature ERP is not a liability. It is critical infrastructure. Production planning rules. Order release conditions. Financial period-close logic. Trade compliance classifications. These are exactly what agents need to call into, not replace. The danger is that most incumbent vendors are not framing it this way. They are defending their business logic as if it is under attack, rather than opening it as a platform for the agents that are coming regardless.

SAP appears to understand this better than most. Rather than fighting the agentic wave, they have embedded Joule, their AI layer, directly into core workflows across finance, HR, supply chain, and procurement. The Cash Management Agent automates reconciliations and is already saving finance teams up to 70% of the time previously spent on manual cash positioning. The Bid Analysis Agent evaluates supplier quotes across unit price, shipping costs, and payment terms, eliminating the spreadsheet work that used to require a procurement analyst. Critically, SAP has added MCP server support and Agent2Agent protocol compatibility, which means Joule can work not just across their own stack but across non-SAP systems. That is a platform play, not a product play. They are not defending their walls. They are opening their gates and positioning SAP data and logic as something external agents will want to call into.

The strategic split

Not all enterprise software faces the same level of threat. Bain’s analysis frames this well. Some workflows are genuine strongholds where AI enhances but doesn’t displace, typically those where data is highly proprietary and switching costs are deep. Some are growth opportunities where the incumbent holds exclusive data and can build the best agentic solution themselves. Some are exposed flanks where third-party agents can hook into poorly protected APIs and siphon margin without the incumbent even noticing. And some are genuine battlegrounds where the incumbent must proactively replace their own SaaS functionality with AI or be replaced by someone else.

The single-purpose SaaS companies, the ones built around one specific workflow rather than a broad platform, sit almost entirely in the exposed and battleground categories. Single purpose apps, a project management tool, a meeting summary app, an expense reporting platform, a list builder, were viable businesses when that one workflow was genuinely hard to build. Now it is exactly the kind of work general-purpose agents automate first. Gartner predicts 35% of these point-product SaaS tools will be replaced by AI agents or absorbed into larger agent ecosystems by 2030. That is not a distant future scenario. That process is already in motion.

The companies in the strongest positions are the ones with the deepest data moats and the most complex domain logic. Healthcare compliance. Financial regulatory reporting. Multi-jurisdiction payroll. Supply chain optimisation across thousands of SKUs and suppliers. These are not problems you can solve by vibing a weekend project. The data is too specific, the rules too complex, the consequences of error too severe.

Workday sits in an uncomfortable position. It manages payroll and HR data for more than 21 million employees globally. That dataset is extraordinarily deep and structurally clean. But the company’s revenue model was built on charging per employee, which creates a perverse alignment problem when agentic AI is simultaneously reducing headcount across its customer base. The company’s answer is its Agent System of Record strategy, positioning Workday as the governance layer for all enterprise AI agents, including those from competitors. It is the right instinct. Whether a company built on managing human headcount can credibly reposition as the infrastructure for replacing it is the question the market is currently trying to price.

Salesforce has moved more decisively. Agentforce ARR hit $800 million in the most recent reporting period, up 169% year-over-year, with 29,000 deals closed. The company has rebuilt its positioning from CRM vendor to what Benioff calls “the operating system for the agentic enterprise.” The pivot is working commercially, though analysts are still pushing for production-scale customer success stories with measurable ROI rather than ARR numbers and deal counts. The underlying move is sound. Salesforce’s data cloud, containing years of customer interaction data across thousands of enterprises, is exactly the kind of asset that gets more valuable as AI needs more context to work effectively.

ServiceNow took a different angle entirely. Rather than fighting to own a specific workflow, they positioned themselves as the control plane for all agents across the enterprise. Their AI Control Tower, combined with a three-year OpenAI partnership announced in January, puts them in the role of the governance layer that every autonomous system running in a large enterprise will need to answer to. Customers report that agents are already cutting sales prep time by nearly 80%. The company is winning enterprise deals not on features but on trust infrastructure.

What this means in practice

There are four moves that separate the companies likely to survive this transition from the ones that will not.

Stop defending the UI. The interaction layer is gone. The energy being spent on better dashboards, improved UX, and polished mobile apps is energy that should be redirected. The question to ask is not “how do we make this easier for a human to use?” but “how do we make this easy for an agent to call?”

Open the data or someone else will open it for you. The MCP protocol has changed the dynamics of enterprise data access permanently. Incumbents that make their data cleanly accessible to external agents, with proper governance and security, will see their platform become more valuable as the agentic ecosystem grows. Incumbents that try to keep their data locked behind proprietary interfaces will find agents routing around them, pulling data via screen scraping, shadow integrations, or simply recommending a switch to a more open competitor.

Shift pricing before the model breaks. Salesforce is inking Agentic Enterprise Licence Agreements at a short-term loss, betting on lifetime value at renewal when the customer is fully embedded. Workday launched Flex Credits to replace headcount-based pricing with consumption-based models. These are the right moves. Any incumbent still primarily pricing on seat count is sitting on a structural time bomb. The businesses deploying AI agents are not adding seats. They are removing them.

Compete on domain depth, not feature breadth. The commoditisation of the UI layer is actually an opportunity for incumbents with deep vertical knowledge. The companies that built a meaningful presence in manufacturing logistics, healthcare administration, or multi-entity financial consolidation did so because those domains are hard. The rules are complex. The edge cases are numerous. The data is sensitive and specific. An AI agent operating in those domains needs exactly that kind of encoded expertise. The opportunity is to become the domain authority that agents need to work with, rather than the system that agents are replacing.

Where we actually are

It is worth being honest about the maturity of this transition. The market signals are real and the strategic direction is clear, but we are still early. The named, quantified enterprise case studies of full ERP replacement by agentic AI do not yet exist at scale. Salesforce has impressive ARR numbers but is still being pressed by analysts to produce production deployments with auditable ROI. The clearest examples of agents replacing professional knowledge workers are coming from adjacent domains. Financial services firms are deploying compliance and analysis agents that are replacing the capacity of dozens of analysts. The evidence is directional but not yet definitive at the ERP level.

That gap is the opportunity, not a reason to wait.

The companies that act now, while the transition is still early, will shape the outcome. They get to define what “agent-ready” looks like for their industry. They get to establish their data and domain logic as the authoritative source that other agents call into. They get to rebuild their pricing model on their own terms rather than in response to a crisis.

The companies that wait for full proof of concept before moving will find themselves negotiating from the customer side of that conversation.

The software market has reinvented itself before. On-premise became cloud. Perpetual licences became subscriptions. Every time, there were incumbents who shaped the transition and incumbents who were shaped by it. The ones who shaped it understood early that the change was structural, not cyclical. They stopped defending what they had built and started building what came next.

That moment is here again. The incumbents that will still be here in 2030 are not the ones bolting AI onto their existing product and calling it an agent. They are the ones accepting that the interaction layer is gone, that the seat model is broken, and competing instead on the one thing that cannot be recreated over a weekend. Thirty years of the right data, governed correctly, in a domain where being wrong is expensive.

That is the moat. Everything else is just a screen.