In Brief
In this Insights piece, we explore the risk artificial intelligence poses to the enterprise-software industry.
The context
Tools like Claude Code and OpenAI Codex, two examples of ‘coding agents,’ are astonishing pieces of technology. They are becoming more impressive by the day. These artificial intelligence ‘agents’ take requests in plain language, propose an approach, run tests, fix what fails and repeat. These types of coding tools appear to be improving much faster than overall AI models. At Generation we are enthusiastic users of these technologies, including for the automation of some software and research operations.
These tools look like a direct challenge to incumbent firms in the enterprise-software space, not only in coding but in broader operations. AI-enabled newcomers allow firms to offer customer service, manage human resources and optimise supply chains. Some customers may even decide to use AI tools to build software themselves. To what extent is AI a risk to the enterprise-software industry?
This article reflects the current state of thinking inside Generation. We have studied the ‘AI and enterprise software’ question in depth, looking at the capabilities of the technology, the incumbents and the new firms in the industry. We have also held more than 50 calls with chief information officers. Our research does not give us certainty – and we welcome feedback and challenge on everything in this piece. On balance, however, we believe that many software firms are well positioned for an AI world.
Why AI is a threat
Generation has long invested in software, including large ‘systems of record’ that sit at the centre of finance, HR, sales operations and supply chains. We have invested in these firms because many of them are great businesses. They have had clear moats i.e., sustainable competitive advantages that other firms struggle to replicate. They have used pricing power appropriately. In our view, they also contribute to a more sustainable world by helping to optimise the use of resources and by making life better for employees.
An enterprise system is more than a storage vault of data. It is also a set of records with meaning, relationships and rules – in other words, information or business logic.1
A simple database entry might say ‘customer_id = 123’ and ‘status = Deal_Won.’ But businesses cannot run on that alone. A real organisation needs the surrounding context, such as which legal entity is the counterparty, whether credit checks have been passed and which internal approvals were recorded. These systems also contain the rules that prevent errors, such as ‘people in role X can see field Y but not Z.’ Crucially, too, writing back to the system of record is how work actually gets done. Without that happening, vendors are not paid and shipping is not ordered.
AI poses a threat from ‘above’ and ‘below’ to the enterprise-software system.
That may sound abstract, so let us explain. ‘Above’ refers to the way in which employees interact with business data. ‘Below’ refers to the data about a company that provides the fuel for an enterprise-software system. Let us take each in turn.
We see several credible threats that operate from above. The first is the rise of AI-native competitors that focus on a narrow, specific use case (although many of these companies have broader ambitions). Big AI companies are increasingly developing their own customised tools aimed at enterprise/professional services. Customer service is a good example. There are now products that pull information from multiple systems, answer ‘where is my order?’, reset passwords, initiate refunds and escalate the remaining complex cases to a human. These tools can replace parts of a customer-service suite, even if they continue to rely on the underlying customer-record system.
A second threat from above is platforms that allow customers to build their own software. AI lowers the barrier to building highly customised applications. Indeed, even someone with zero expertise can use Claude Code to create a basic bit of software in a few hours. If self-built software becomes more popular, demand for packaged software may decline.
A third threat from above is control of the interface. If a company chooses a single assistant as the front door to work, that assistant can become the place where employees spend time, ask questions and initiate actions. For instance, you as an employee might do all your work through Google’s Gemini, even if that Gemini interface draws on other apps to get things done. When that happens, the enterprise-software vendors can lose visibility. This makes them more vulnerable to disruption, as value shifts toward the layer that owns the interface and away from the system that stores the data.
The threats from above pose a number of key risks for enterprise software firms. As agents take over human work, they may reduce the number of seat-based subscriptions. Existing software is built to reflect a human-centric model, and while humans will always be crucial, the relative role of AI agents will force a rethink of many basic assumptions. And by taking the user away from enterprise apps, AI risks relegating them to a dumb back-end that hosts the data. If enterprise-software firms are viewed as mere data depositories, they may have little pricing power. Very few people think that AI will put the major software companies completely out of business. The question is whether their growth prospects are threatened.
Over time, there is another risk. The AI challengers may start to understand the business logic inherent in the vault of data, negating the need for the enterprise-software firms. This is likeliest when models get broad, repeated access to the metadata: the schemas, role rules, workflows, approval graphs and long histories that encode how a firm actually runs.
Alongside these pressures from above, there are pressures from below. As we have shown, software applications require databases in order to work. The companies running those databases (including companies such as Microsoft and Amazon) have a clear incentive to create AI tools that also learn the business logic contained in that data. In that world, the enterprise app system of record becomes unnecessary.
All these threats are serious. And while AI products from software incumbents are nascent, startups have proven to be quick off the mark. Our internal estimates suggest that AI-native challengers already have several billion dollars of annually recurring revenue. By contrast, the dedicated AI revenue lines disclosed by large incumbent enterprise-software vendors remain – for now – closer to around $1 billion in aggregate (overall enterprise-software revenues, of course, are orders of magnitude larger).
Our view
How threatened is enterprise software? There are different ways of answering this question. The first step is to assess how much AI has truly changed the ‘buy vs build’ calculus of enterprise software customers. Will enterprises build more software in-house with the use of AI?
We believe it is unlikely, at scale. At most companies, wages are the biggest single cost. By comparison, software spend is tiny. Across the US economy as a whole, total spending on employee compensation is more than 20 times total spending on software.2 That has an important implication: once you consider not just the cost of building, but also of maintaining and upgrading software, it rarely looks cheaper to do it internally than to buy it from a dedicated provider.
AI reduces the cost of producing the first version of a tool. It does not remove the need to secure it, maintain it, update it when upstream systems change, document it for audit and support it when employees leave. We do expect more internal build activity at the edges, especially in areas that are strategically important for a business, such as underwriting for an insurer. For non-strategic domains such as HR, however, we continue to hear a preference for buying and customising.
We are mindful that this trend could change. At some point, coding agents may also be able to maintain software systems as well as to create them. But that could be some time away. In any case, our survey work suggests that the share of companies choosing ‘build’ over ‘buy’ is currently declining, rather than rising. In addition, it is worth remembering that the division of labour is an extremely powerful force. Efficient companies outsource activities in which they do not have a competitive advantage. Over recent decades, companies have outsourced more and more software spend, in part because packaged software has become better at meeting needs that are common to different types of businesses.3 There are compelling economic reasons to expect that trend to continue.
The bigger question, we believe, is whether enterprise-software firms can capture the opportunity from AI. To begin to answer this question, let us go back to first principles. Generation has invested in technology for more than 20 years. Over this period, numerous developments have threatened the leaders in this industry. In practice, however, the number of clear, permanent displacements was small. We often cite IBM as the only clear example of a transition away from a previously dominant position in enterprise computing.
Two factors explain this phenomenon of ‘many threats, few displacements’ in software. The first is ‘ecosystems and distribution’. Large tech firms invest heavily in security, reliability, regulatory updates and global support. For customers, the cost of switching away can be significant. The second is that the better incumbents have learned from prior technology shifts. They have read the disruption playbooks. They have built product teams that can ship fast.
In our view, enterprise software firms recognise the scale of the AI challenge they face, and are investing accordingly.
First they must decide how they interact with AI. At one extreme, they can close their data off from AI agents entirely. This is tempting, as it reduces competition today. But it increases the risk of newcomers developing products to route around them. At the other extreme, they can share their data freely. This can help develop partnerships today, but runs the risk of AI learning about business logic.
They are, correctly in our view, choosing a middle ground: sharing data but protecting context. Software firms can make it easy to read from them, ensuring that AI apps want their data, but they are trying to make it hard for AI to replicate their full business logic. For instance, an AI agent might be able to see that a company has offered a client a 10% discount, but it would not know why that particular number was chosen. The goal is to expose actions (‘create order’, ‘approve discount’) and answers (‘approved discount is 10%’) rather than the underlying logic that produces them. If AI remains fenced off from the data and relationships inherent in an enterprise software system, then they are fenced off from the very thing that makes enterprise software so valuable.
The second step involves enterprise software firms developing their own AI capabilities, either by internal development or through acquisitions. This can expand what software systems do, including how many processes can be automated. It can also strengthen the value of having a single place where rules are enforced. On top of this, relying less heavily on seat-based pricing, and more heavily on outcome-based pricing, can alleviate some of the risks involved in AI agents taking over from human agents – even if the eventual effect on unit economics is highly uncertain.
Many enterprise-software firms are well positioned to develop AI capabilities. They can use their great scale and deep data to improve the reliability of AI automation. They can also package AI features in a way that reduces the need for customers to stitch together a long list of products. These firms have deep pockets to fund AI innovation. In recent years vendors have improved their financial performance, in part through judicious cost control. Together Salesforce, SAP, ServiceNow and Workday generated almost $30 billion in free cashflow last year.
We see evidence of AI innovations across the large vendors we follow, meaning that AI revenue growth should soon rise. The more threatened apps are responding best. Many have embedded AI assistants into their existing workflows, so employees can ask questions, generate drafts and initiate actions within the existing control framework. Several are building role-focused assistants that take a specific process, such as resolving an IT ticket, and automate the steps under supervision.
In our view, the development of AI capability can accelerate the use of software.
Natural-language interaction can make such systems easier to use. AI can make configuration easier by suggesting how a process should be set up. It can automate data entry and reconciliation, which can improve the return on a software investment. Demand for reliable systems that hold records and enforce rules can grow alongside the growth in AI. Recently, Nvidia chief executive Jensen Huang said the idea that AI will replace software tools was “illogical,” and he believes that companies will continue to rely on existing software rather than rebuild tools from scratch.
We would also add that AI will have implications for software firms’ costs, as well as their revenues. AI is very good at automating software development (we hear of some companies growing their developer productivity at extraordinary rates), as well as other white-collar jobs. Software firms, whose biggest cost is labour, stand to gain.
Concluding thoughts
All of this leaves us with the following view. We think enterprise software remains central in an AI world. We think the best-positioned companies are those that control business meaning, enforce rules and provide a safe place for AI systems to take action. We are watching for evidence that automation leads to widespread reductions in paid users. We are watching for evidence of internal builds at scale. And we are watching for evidence of agents becoming better at understanding business logic. Finally, as well as threats from above and below there may be threats from the side. AI can enable existing companies, including large incumbents, to more easily move into areas adjacent to their current areas of operation. The result could be consolidation within the industry. If those signals strengthen, our view will evolve.
We acknowledge that there are risks. There is tremendous, perhaps unprecedented, uncertainty about how AI develops. This could be a technological shift like no other.
But we are also mindful of how the market is currently pricing these risks. We see material upside in our software names, even assuming fairly conservative forecasts. If we ask ‘what is the market pricing in?’, we see scenarios that would be unprecedented in the history of technology. For example, Workday’s shares are currently pricing in a scenario where they exhaust their current contracted backlog, grow by just 0-2% beyond that point, and see modest margin expansion as sales and marketing spend falls. Overall, markets are currently treating software companies monolithically, and essentially valuing them all as mere databases. We think that is overly pessimistic.
We would welcome your challenge, especially where you believe the substitution risk is highest. Our current judgment, grounded in our internal research, is that AI changes the shape of enterprise software but not the need for it. The systems that keep official records, enforce permissions and permit actions look likely to remain crucial. The vendors that run those systems appear to be responding through product changes and acquisitions that keep them embedded in the new workflows. If that response continues, AI may expand what software can do.
- To be clear, software firms are not the sole repositories of business logic. Much of that exists in other forms, including email and Excel spreadsheets, as well as in ‘tacit knowledge’ that is not codified anywhere.
- U.S. Bureau of Economic Analysis data on national income
- U.S. Bureau of Economic Analysis data on private fixed investment in intellectual property products. In 2024, less than 15% of US software spending was on “in-house” software, down from more than 45% in the 1980s.
Important information
The Insights: Our thinking on enterprise software in a world of AI report is prepared by Generation Investment Management LLP (“Generation”) for discussion purposes only. It reflects the views of Generation as of February 2026. It is not to be reproduced or copied or made available to others without the consent of Generation. The information presented herein is intended to reflect Generation’s present thoughts on sustainable investment and related topics and should not be construed as investment research, advice or the making of any recommendation in respect of any particular company. It is not marketing material or a financial promotion. Certain companies may be referenced as illustrative of a particular field of economic endeavour and will not have been subject to Generation’s investment process. References to any companies must not be construed as a recommendation to buy or sell securities of such companies. To the extent such companies are investments undertaken by Generation, they will form part of a broader portfolio of companies and are discussed solely to be illustrative of Generation’s broader investment thesis. There is no warranty that investment in these companies have been profitable or will be profitable. While the data is from sources Generation believes to be reliable, Generation makes no representation as to the completeness or accuracy of the data. We shall not be responsible for amending, correcting or updating any information or opinions contained herein, and we accept no liability for loss arising from the use of the material.