At Scope 2026, Mikael Johnsson opened with the question that’s been hanging over B2B software for the past year: is AI killing software? The short answer is no. The longer answer is more interesting.
A glimpse of what took place at Scope 2026.
The headlines vs. the data
Public software valuations are down: In early May median EV/revenue multiples sat roughly 58% below the ten-year median, the lowest in nearly a decade. If you were only reading the headlines, you’d be forgiven for thinking the entire asset class is in terminal decline. However, as of the last few weeks, there’s been a dramatic reversal with the Bessemer Cloud Index up significantly and with individual companies like Datadog and Snowflake up more than 100% in two months. I would argue that three major forces were driving the sell-off, none of which was based on any real fundamental analysis and all of them now getting de-bunked bit by bit.
First, there’s a limitless fascination with generative AI, and it’s not hard to understand why. Four-year-old startups are being acquired at $60 billion valuations. Companies are reaching hundreds of millions in ARR within months of launching. A startup raised a $2 billion pre-seed round before shipping a product. The excitement is real, but it’s creating a halo effect where capital is (was?) flowing indiscriminately toward the new and away from the existing, regardless of underlying business quality and long-term sustainability.
The excitement is real, but it’s creating a halo effect where capital is (was?) flowing indiscriminately toward the new and away from the existing, regardless of underlying business quality and long-term sustainability.
Second, there’s a massive oversimplification in how the market categorises software. Not enough discrimination is being applied between companies that are AI net positive – with a genuine opportunity to build stronger businesses – and those that are AI neutral or AI net negative. The nuance matters enormously, and this realisation is what’s now driving the rebound in stock prices for Datadog, Snowflake and the likes.
Third, liquidity flows in capital markets are just that – liquidity-driven. Once someone starts dumping a software name, passive index funds follow, and the vicious cycle takes hold. It’s a momentum trade, not an analytical one. And the same, of course, is true on the upside; once a stock starts appreciating, passive liquidity flows follow and reinforce the momentum.
The picture is uneven – and that’s the point
Apply a finer lens and the sell-off tells a more nuanced story. Let’s look at what happened since GPT-5 was launched. When I looked at the data in early May for our annual Scope Summit, vertical software was up 3%, infrastructure was up 2% — but horizontal software was down 35%. The market was punishing major software names indiscriminately, but under the surface the divergence is (was?) stark: companies growing above 20% year-on-year still trade at 8x forward revenue, while slower growers have compressed to 1–3x.
This matters because the bear case for software isn’t wrong, it’s incomplete. There are genuine structural pressures: AI-native competitors are emerging fast, per-seat pricing is under threat, and some incumbents’ AI transformations are more “press release” than product reality. But the bear case treats software as a single category, and it isn’t. Vertical software has protective moats through proprietary data, regulatory complexity, and deep process embedding. And much of infrastructure software is experiencing strong AI tailwinds as the amount of data that moves around increases dramatically. Horizontal “shallow workflow” tools are the ones facing a genuine reckoning. While the markets make the distinction, the headlines don’t, so the discourse lacks nuance.
The complex, mission-critical systems that underpin our economy in places like banking, insurance, healthcare, and energy are not going to be vibe-coded anytime soon. There’s just too much complexity, rules, legislation, governance and compliance inherent in these systems. The same goes for the major systems of record that exist in all organisations – things like ERP, CRM and HCM. AI agents need these systems and their data to function, for example, an accounts payable agent that can’t access the general ledger hits a roadblock immediately. An automated stock replenishment agent that can’t access the major inventory system faces the same fate. Also, many incumbents aren’t sitting still; they’re undergoing real AI transformations, backed by brand and distribution advantages that most startups can’t match. That said, systems of record will increasingly struggle to remain as the systems of interaction, as agentic interfaces change how users engage with applications. Their data will be essential; their grip on the user experience will loosen. The ones who will successfully navigate this transition will move from a per user, per module pricing model towards one based on data access and agentic transactions executed on their platform.
The explosion
So if AI isn’t killing software, what is it doing? It’s creating more of it than we’ve ever seen.
Coding agents are democratising software development. The pool of potential software entrepreneurs is expanding from roughly 50 million professional developers worldwide to something dramatically larger, and the cost of building is collapsing alongside it, making it viable to create products for a thousand users, a hundred, or even one. The result is an explosion in the number of software products, companies, and use cases.
Meanwhile, autonomous AI agents are becoming the new software users, consuming applications around the clock, extending automation into business processes that were previously human-only territory. They don’t need a lunch break, a weekend, or a four-week holiday. They’re always on. The addressable user base for software is growing exponentially — and so is the usage.
This isn’t speculative. Battery’s data on enterprise AI adoption shows that after years of expectations running ahead of reality, the gap has closed and in Q3 2025, actual adoption surpassed projections for the first time. The number of identified gen AI use cases within enterprises is exploding, roughly doubling quarter on quarter, with the largest organisations identifying 250+ use cases and nearly half of them already under development.
From a deck presented at our Scope Summit 2026. Source: Battery. Adoption data — actual vs. projected gen AI adoption over time, showing gap closing and inverting.
From a deck presented at our Scope Summit 2026. Source: Battery. Total number of identified gen AI use cases by budget range + status of adoption.
Coding agents led the way. Marketing, sales, and finance are next — customer support is moving from chatbots to fully agentic solutions with a significantly broader remit and ability to autonomously solve complex support use cases, large parts of traditional SDR and BDR processes are being automated to drive pipeline creation at scale, AR and AP reconciliation, FP&A forecasting, close and consolidation are all ripe for agentic automation. Supply chain, HR, and field service are following.
The context problem — and a new category of software
Most of the AI agents deployed today are still relatively simple — a copilot here, an automation there, each focused on a discrete task with clear instructions. But as agents move into complex business processes with multiple stakeholders, imperfect information, and real consequences, they hit a fundamental bottleneck: context.
Today, humans are the glue that makes traditional software work. We input data. We fill gaps. We make judgement calls, exceptions, executive overrides. We carry the contextual knowledge that no system currently captures, things like: we always give this customer 15 extra days on payment terms because we know they pay on time. Or: we always ship from this distribution centre for that client, even though it’s further away, because it lets them consolidate deliveries. Or: this customer always pays late and in arrears, so we adjust our liquidity forecast accordingly.
These are some of the thousands of contextual decisions being made daily across every organisation. They sit in people’s heads, or at best in an email thread, a Slack message, a call transcript. They’re not in the ERP, they’re not in the CRM or any other system of record. And without them, an agent trying to automate a complex process will rapidly hit a wall.
This is where the context graph comes in, and we think it’s one of the most important emerging concepts in enterprise software. Traditional systems of record are state machines — they capture what is happening and what happened. The context graph captures why: the decision traces behind every action. Why did we provide that discount? Why did we replace that product? Why did we capitalise that expense? It draws on structured and unstructured data sources and builds incrementally over time as agents capture decision traces at runtime. Eventually, the context graph becomes something an agent can invoke to make autonomous decisions based on the right context, not just the right data.
This is still an early category. The logic is compelling and the need is becoming urgent as agents move from simple copilots into complex, multi-stakeholder workflows. For founders and operators thinking about where defensible value will accrue in the agentic era, this is a space worth watching closely.
What the durable winners will look like
Witht the advent of generative AI we’re transitioning from the SaaS-era to the Enterprise AI- era. The playbook for building a software company has changed. MVP timelines have collapsed from months to weeks. Three to five engineers can now ship what previously took twenty. Development teams are reconfiguring: separate roles for backend, frontend, product, UX, and QA are collapsing into smaller squads where everyone is a builder.
Pricing is shifting too. Per-seat models that served SaaS well for two decades make less sense in an agentic world. The question is no longer “how many people can use this product?” but “what work actually got done?” — how many support tickets were resolved, how many loan applications processed, how many forecasts run. Outcome-based pricing aligns the software company’s incentives with the value it creates. The transition won’t be instant and hybrid models will persist, but the direction is clear.
The companies that survive this Cambrian window will share a few major traits:
- they’ll be deeply embedded into core business processes
- they’ll compound value through proprietary data and context; and
- they’ll price on outcomes rather than seats.
For founders, that means building depth and defensibility from the start. For operators inside existing companies, it means understanding which side of the vertical/horizontal divide you’re on, and acting accordingly. For investors, it means applying the kind of sector-specific diligence that separates the AI net positive from the AI net negative, company by company, rather than treating the category as a monolith.
More builders. More products. More users. More usage. Entirely new categories. That’s software’s Cambrian Moment as we move from the SaaS-era to the Enterprise AI-era. The companies capturing the opportunities in this golden-moment will become the next generation of global software leaders. We are convinced that a meaningful number of them will be European and our job as investors in European Enterprise AI companies is to help them improve the odds of that being the case.
