Found something good?

One of the biggest topics right now in the tech world is what does the future of software look like when AI agents are doing a substantial amount of work in the enterprise. Does software get compressed and simply become a database that agents leverage? Do we use AI agents to code all our own personalized enterprise software in the future? Do startups or incumbents win this new battle? How big are software markets in a world of agents using these tools? And much much more.

It’s impossible to try and tackle all of these questions sufficiently in one sitting, but I figured I’d lay out a bit of a roadmap for what enterprise software looks like in the future.

Why companies buy software

Firstly, we should all get on the same page about why companies use and buy enterprise software in the first place. You can think of enterprise software as a codification of a company’s processes. These are the parts of your company that you don’t want to change spontaneously, because they provide the structure for your workflows and operations to keep your business running, or allow you to do the other work that really matters.

A large car company decides to procure an ERP system because it’s the backbone for complex global operations where failures could mean far more than the cost of the system. It’s responsible for keeping track of tens or hundreds of billions in revenue, and the company wants to ensure that every single transaction that they make goes through the exact same way every single time, with five 9s of uptime and reliability, and with an output that they can get accepted and approved by the auditors. Jobs are on the line, all the way up to the CEO, based on how well this system performs.

And the reason that companies choose to use software vendors that specialize in certain domains instead of just building this technology themselves can best be understood by the concept of “core” vs. “context,” popularized by Geoffrey Moore. The core stuff is what makes your company unique (like your core product, your talent, your culture, and so on), and the context is absolutely necessary, but largely non-differentiating between firms (think all of your systems of record like payroll, IT ticket responses, ERP systems, content management platforms, and so on).

For software that handles your company’s context activities, your customer will rarely notice when these things work great, but they’ll certainly notice the downstream impacts when they don’t. This is why you “rent” these services from software providers that do this as their day job for thousands or millions of other companies. And incidentally, much of this software helps define these underlying workflows in your organization so you don’t have to -- so each company doesn’t have to reinvent the wheel each on how to handle HR, IT ticketing, and so on.

Deterministic vs. non-deterministic systems

Ok, but if an enterprise is filled with 100X more AI Agents in the future, won’t that eventually consume all the use-cases that software did previously?

If you take the analogy of the industrial world, software is like the machinery in a manufacturing plant; and agents are like the people that operate that machinery. Just like machinery in a plant, in deterministic software, you want things to happen the same way, every time: you want data that goes into your ERP system to be calculated based on the business rules you’ve setup; you want security permissions you’ve set for accessing critical files to behave the same way every time, and not leak information between users; and you don’t want an HR system probabilistically generating new reporting structures for employees on every load.

Conversely, you want *non-deterministic* systems (in this case AI agents) for the things that people have always tended to do in these systems. An agent filling out notes from a user’s client meeting in a CRM system, generating a sales presentation for a new customer, answering a customer support inquiry with the best products to leverage, doing research or providing analysis on a set of enterprise documents and data for an important decision, and so on. As a result, the non-deterministic and deterministic elements of software end up working in a complementary fashion.

In contrast to some of the public discussion, I’d argue that in a world of 100X more AI agents than people in an enterprise, the value of the systems of record and tools agents will use will go up, not down. Because in this new world, software provides the guardrails on which agents can operate successfully within an enterprise, and gives them the underlying tools to use to be more productive themselves and work alongside people.

For instance, if you ask an AI agent to generate 10X the number of outbound email and marketing messages, where will all those agents find the data to work with, and where will the leads end up going to have a human sales rep act on them? Almost certainly some form of a CRM system. Or when an AI Agent processes reviews invoices and transactions, where will they enter this data when they’re done to kick off a supply chain workflow? Again, an existing or reinvented ERP system. As long as we have humans and agents interacting with one another, there will need to be deterministic software that bridges these interactions seamlessly.

And as a result of all of these new agentic users within our software systems, the size of enterprise software markets are going to increase massively.

Unconstrained software markets

Traditionally, software companies have been stuck within the constraints of existing IT budgets, which tend to tap out around 3-7% of a company’s revenue. This creates an inherent upper limit for what the budget can be for software, which translates into the total addressable market of various technology categories. Now, with AI Agents, the software is actually bringing along the work with the software, which means the budget software players are going after is the total spend that goes into doing that work in the company, not just the tech to enable it. This inevitably leads to a substantial increase in TAM for most software categories whose markets were artificially held back in size previously.

For instance, a contract management vendor just a few years ago was limited by the number of attorneys within an organization that needed to manage and review contracts. However, in a world of AI agents, that cap no longer exists, where you might have agents running continuously processing and reviewing contracts and providing extend legal services for any company. The size of the legal services market in the US is somewhere around $400B, nearly 50-100X larger than the corresponding software categories in legal space. You can easily see how AI agents making the legal industry meaningfully more productive could capture multiple percentage points of that $400B in spend, thus multiplying the size of the legal software market overnight.

The same analogy holds for almost any software category that was artificially constrained by a customer's headcount, or limited software spend, previously. We're already seeing this in spaces like coding, where the AI agents in coding are generating orders of magnitude more revenue for the new coding startups than the prior IDE categories generated. And we can expect this same dynamic for most other areas of knowledge work over time, from healthcare and financial services to legal and marketing. Tools that bring along work will generate far more revenue than the prior era of tools used to.

The opportunity ahead for software

Now, the big open question is who wins in this technology wave. If we have learned anything from Clayton Christensen’s Innovator’s Dilemma framework, we know that incumbents tend to lose in markets when there’s a new tech or business model innovation that is unattractive to them because it represents less revenue or profitability. Conversely, they tend to hold their own when the new business model or technology is supportive of their existing business, or perhaps even enhances it. Ultimately, in AI, we’ll likely see examples of both outcomes.

Some existing software categories, like customer support or coding, are changing so rapidly that many incumbents will be unable to change their business models or technology stacks fast enough to respond to the growing demands of the new market, even if getting to the other side would be economically attractive. Similarly, as Jaya Gupta and Ashu Garg outlined in their piece on context graphs, there are many new categories of agent-native software that will need to be created for the first time because no existing vendor logically can capture the way the agent needs to do its work. This will create tremendous opportunities for new startups to build AI agents within these categories, and we’re seeing new ones crop up every day.

Conversely, there are a number of existing SaaS players are in a natural position to power agents in their relevant work categories, especially those that have the natural context necessary for AI agents to be successful in automating work. Unsurprisingly, I’d expect there to be a correlation between the strength of a software company’s moat and the amount of data they house, how complex the underlying workflows are, the number of connections there are with other systems, how intertwined a human-in-the-loop element is to the underlying agentic workflow, and how naturally AI agents can be plugged into these workflows (either as native users or via API in external agentic products).

Of course, this will mean that the business model of software (new upstarts and incumbents) will have to evolve over time. Today, the vast majority of SaaS products charge on a per-seat basis, which generally corresponds to most of the usage that the software sees today by its end users. But in a world where AI agents do most of the interaction and work on software, enterprise systems will have to evolve to support more of a consumption and usage-based model over time. AI agents don't cleanly fit as seats on software, because any given AI agent can do a varied amount of work within a system (e.g. you could have 1 agent doing a billion things or a billion agents doing one thing).

Inevitably, we’ll see a mix of both outcomes. Users (or people) will continue to be seats within software, and AI agents that represent extended usage will be monetized through consumption. We're already seeing this play out in AI-native coding products and other platforms where there's an end-user seat component mixed with a consumption model on top. Obviously depending on the human vs. agent value proposition mix, the revenue stream will slide between these two ends of the continuum.

Ultimately, there’s still plenty that needs to play out before we fully see what the future holds for enterprise software. But it will certainly be the most dynamic and exciting period we've ever seen in software history.