Whilst tongue-in-cheek, I’ve always found the advice ‘never make predictions, especially about the future’ to be solid, and never more so than in AI-land.
Three years ago, AI was touted as an accelerant for modern, cloud-native software companies, super-charging well-staffed, industry-leading developer teams to deliver better products at an increasingly accelerated rate.
They were hot property, and valuations reflected it.
Fast-forward to today, and AI coding assistants– led by Anthropic’s Claude Code – have flipped this narrative on its head.
SaaS companies have gone from being AI’s predicted winners to those at most existential risk.
The AI threat
If AI means anyone can code, and AI agents can work 24/7, then surely it is just a matter of time until even leading software products become a cheap, generic commodity?
Cue the ‘SaaSpocalypse’, with public equity markets seeing a collapse in share prices of some of the world’s biggest names, as AI-induced panic took hold.
Unfortunately for Private Equity and Asset Managers, this coincided almost precisely with growing concerns that net asset valuations were already overstated, with worries from shareholders, regulators, and governments that insufficiently robust valuation, revaluation, review, and due diligence practices were creating valuation bubbles.
There has since been a strong and well-publicised push-back from within the industry, but the anxiety nevertheless remains, and we’ve received a flurry of demand to evaluate individual assets and whole portfolios for potential exposure to AI-driven disintermediation.
So what are we to make of the AI kills SaaS hypothesis?
Well, the risk is real.
The replacement hypothesis
Generative AI, and the LLMs that underpin it, have by some measures already exceeded human-level coding capabilities.
It’s been an area of focus for frontier labs for three reasons:
- Coding has all the characteristics of a perfect AI use case:
– A clear, well-defined problem to be solved
– A textual, logical, rules-based output
– A well-structured, publicly available, high-volume training data set (GitHub, Stackoverflow etc.) - In the short term, it enables rapid development and shipping of new features, as best evidenced by Anthropic’s unparalleled Q1 release schedule
- Long-term, it provides the basis for ‘recursive self-improvement’, whereby each generation of AI trains its successor, exponentially increasing the rate of progress
AI coding, leveraged in a broader agentic workflow – one agent writes the requirement, another writes the code, another QAs, another tests, all overseen by an orchestration agent and with stage-gated human-in-the-loop approvals – could well be the entry-point into a brave new world where time-to-market for software collapses, drastically increasing competitive market pressures and pushing out incumbents.
Case closed. Software is dead. Long live software.
But of course, and as ever, things aren’t quite so simple.
Whilst directionally correct, I believe this case is overstated for several reasons.
First, a software business is more than its codebase. Its brand, expertise, experience, people, customers, contracts, and data are all powerful value-drivers that AI can’t replicate, and leveraged well can create powerful defensive moats that all but the most well-funded startup will struggle to overcome.
Second, it’s one thing to develop a better product; it’s quite another to displace the one you’ve beaten. Particularly in an enterprise environment, where multi-year contracts, risk-averse buyers, and checklist and reference-driven procurement are the norm, a challenger product must be significantly more compelling than an incumbent to win the day, and a healthy dose of luck is required.
Third, human incentives. Just because AI-assisted coding can accelerate the development of a competing product, it doesn’t mean somebody will. In reality, there’s a finite number of entrepreneurs to go around, and some markets will simply be seen as too difficult, or not valuable enough, to disrupt.
Fourth, and finally, this hypothesis assumes incumbents are standing still, awaiting their inevitable demise. In reality, software companies are some of the most aggressive and advanced adopters of AI, using it to rapidly refine and adapt their own product whilst their challengers play catch-up.
The pricing disruption dynamic
The second major concern is focused less on wholesale displacement and more on fundamental challenges to the SaaS business model, manifesting on two fronts:
- Price compression: As new, AI-native entrants proliferate, a pricing ‘race to the bottom’ ensues, as increasingly crowded marketplaces become characterised by intense competition
- Seat-based pricing: As AI disrupts the job market more broadly, AI-induced enterprise headcount reductions (or at least stagnation) will put pressure on traditional SaaS seat-based pricing revenue models
To my mind, there’s no doubt that these dynamics will manifest.
But not equally, not everywhere, and not all at once.
Simple, commoditised software products where development timelines and barriers to entry are low are no doubt at risk of severe price compression in relatively short order. Conversely, complex products in highly-regulated environments are likely to be heavily insulated, at least in the short to medium term.
Seat-based pricing pressure will also manifest, but only where customer seats are truly at risk. Whilst AI legitimately threatens a significant proportion of the workforce, not all occupations are equally at risk, and SaaS companies serving better-protected professions or industries will remain largely insulated.
Furthermore, if SaaS companies can transition to consumption-based or outcome-based pricing models, they may be able to further shield themselves from this dynamic.
Evaluating your exposure
So, if you’re wondering whether AI rings the death knell of your software portfolio, the answer is… maybe.
The reality is that exposure varies enormously based on the specifics of the market and the individual organisations within it, and whilst we’ll no doubt see ‘blood in the water’ as companies with heavy exposure go to the wall, we’ll see other, better-protected incumbents survive or even thrive in the AI era.
Ultimately, AI is not an executioner, and the ‘SaaSpocalypse’ doesn’t spell the end for software investing – but it does require new working practices adapted to the AI age:
- Evaluations of existing portfolios to determine point-in-time AI Risk exposure
- Updates to due diligence and portfolio management processes to include explicit consideration of AI risk
- Ensuring existing assets and investments are integrating AI into their operations and business models
- Advising portfolio companies on the key defensive actions to take
So, my prediction?
Well, nobody can know for sure what the future holds. But what we can do is prepare ourselves to face it.
By integrating AI risk into portfolio and investment management processes now, forward-thinking companies can protect their downside, whilst positioning themselves to capitalise on the next AI-led generation of SaaS winners.
