Following our AGM in November 2025, which included presentations on our activities in AI, I am often asked for an updated perspective based on recent developments in the technology and public markets. Below is a focused summary of what we see happening and our take on what is likely to evolve.

To my mind, there have been three major events in the months since our AGM:

The model releases in Q4 across Anthropic, OpenAI and Google showed a significant (and somewhat unexpected) step change in performance for business applications. That capability will filter into applications over time, but it emphasised again the sustained capability of AI to impact white collar work over time;

The momentum of many key players in AI accelerated – January 2026 capped a year of huge growth for the platforms, various application companies and access to funding. We can expect this to continue;

The stock market is taking a much more bearish view of multiples of classic SaaS companies. Companies that in most cases are largely still hitting current profit forecasts.

The change in market sentiment might result from macro factors, tech sector rotation, or an assessment of more structural changes. While it is hard (and perhaps ill advised) for me to comment on the sentiment of public market investors, I can address possible structural changes and how a long-term software investor (like Hg) views the outlook for application software today and its long-term potential.

Our view from up close

History may come to suggest that late 2025 was the period that AI went enterprise. The early excitement of OpenAI and steep consumer adoption curves that we all saw in 2024 were superseded by the focus on ARR / EV growth in Anthropic, the product market fit of AI led applications in coding, customer support, and legal, as well as an increasing sense of the potential within business applications. The elephant in the room question essentially became ‘at what pace’?

At Hg we have seen a similar dynamic.

We have two decades of experience investing in software businesses and today we are investors and board members in nearly 60 privately-owned software businesses with an enterprise value of more than $185bn combined. And as discussed at prior AGMs, the emergence of previous general-purpose technologies and its diffusion into the real economy takes time. With AI, it feels like everything is happening everywhere; but it will not happen all at once.

What is clear across at least 1,400 operational AI projects and 100+ AI product builds is that AI products are gaining real traction with customers. That adoption is happening, bookings are real, and customer ROI is evident. And that is taking spend from the traditional sources of software growth – new seats, upsell, and cross sell. As we have flagged before, traditional SaaS revenue growth rates are slowing (by perhaps 3-5% in last 12 months) and the driver is a pick-up in AI spend from end customers.

It is also very evident that once you get deep into an application – into regulated sectors, into professional led advisory, into SME and mid-size enterprises across Europe without huge IT budgets, that AI adoption will be measured and likely led by incumbents.

We discussed at the AGM that incumbent businesses will win if they mobilise - by leveraging the ‘four Ds’ characteristics of deep vertical workflow automation:

Data: proprietary datasets such as tax filings, regulatory records, transaction histories and the associated workflows to produce.

Domain: deep vertical-specific knowledge accrued over decades embedded into the technology.

Distribution: established, trusted customer relationships. Brand.

Deterministic: mission-critical processes that require accurate, predictable outcomes (not probabilistic responses).

In other words, these are companies that have spent years mastering a difficult workflow, using a complicated set of rules and processes combined with proprietary data and deep integrations, both up and down stream. Together, this provides a huge advantage when building agentic products on top.

From systems of record to systems of action

The commercial opportunity for those that mobilise is huge.

Until now, most business software has been some kind of ‘system of record’, storing data and tracking activity (“automating filing cabinets” as some have described it in the podcasts).

GenAI, in particular agentic solutions, now enables the incumbent software providers to become 'systems of action’ where agentic Gen-AI software can manage parts of the workflow itself. Increasingly, that means small teams of experts directing a set of agents that can execute bounded tasks end-to-end under human oversight. As products take on more of the work, they can also pull ‘edge’ tasks into the software itself. This enlarges the target addressable market (TAM) in two ways. First, there is human activity now captured by the product and lower delivery friction opens up lower value customers and long-tail segments that were previously uneconomic to serve. This is where we see TAM expansion for AI led software – accessing labour and increasing scope.

Today, the clearest example of this is software engineering itself. For many years engineers used software to manage, store and track codebases. Today, agentic AI ‘engineers’ are writing the code, doing the work of junior and mid-level engineers, with senior engineers overseeing their output. And the volume of code produced is perhaps 3x what it was before – tackling micro niches and needs.

It is by now familiar also to hear the view that this opportunity is huge. Software firms no longer aiming at the $1 trillion of annual software spend but the $50 trillion-plus human labour market. Let’s not get carried away, but incumbent SaaS businesses will capture some of this opportunity if they ‘re-found’ themselves as ‘AI-first’ - and quickly. This won’t be simple, but we have been creating and cross sharing repeatable playbooks for a few years to maximise the opportunity from this capability. In this context, value accrues where decisions and actions happen, not just where they’re recorded.

The unit of value will move too. Pricing will start to reflect outcomes such as tasks completed, or labour time saved – although this will take time. I haven’t spoken to many AI first CEOs or founders who say that they have cracked pricing models in markets where per seat has prevailed for so long. But they intend to. And some are now accessing labour budgets within their customers.

The initial data from our portfolio also shows that margins from AI ‘products of action’ are similar to SaaS ‘products of record’ – with consumption-based models not proving as popular as some anticipated due to the spend uncertainty they create and there being very effective means to manage compute costs.

How to go AI-first

We are working with many of our forward-looking portfolio companies on two fronts – product and operations - to help them ‘re-found’ themselves and become AI-first.

On Product: Hg Catalyst is our AI product incubator. It deploys small, senior teams to sit inside portfolio companies and accelerate AI product builds. The focus is on defining narrow workflows with clear success metrics, fast release cycles, and reuse of patterns so each new build is quicker and safer than the last. We operate across the portfolio with a scaled capacity of 80+ engineers, product managers, and designers. This also means a breakthrough in one business can be rolled across the rest of the portfolio in weeks, not quarters.

Second is operational AI transformation. This means working closely with our leadership teams and functional heads to reimagine and reshape how their Engineering, Customer Support, Finance, HR, and Legal teams operate in an AI-first world. We are targeting step changes in output. That means better products, more products, faster cycle times, better customer experience and clearer audit trails. And margin available to re-invest in product.

Partnerships and platforms support both tracks. We are working closely with the leading AI labs and builders, keep our architectures model agnostic, and creating shared components for data pipelines, permissions, telemetry and observability. This allows teams to focus on the last mile and preserves flexibility as model economics and performance move.

GTreasury, the treasury solutions platform acquired by Ripple last year, is a good example. Our Hg Catalyst team worked closely with the company to develop GSmart AI, a new agentic product that proactively identifies risks and variances and recommends strategic actions for finance leaders. Previously these were tasks that a human would do after exporting and analysing data from the platform. Now it’s all done by agentic AI software - and customers love it. But that’s because it’s built on top of decades of data and deep domain knowledge.

We plan the same kind of transformation with Onestream Inc, a strategic acquisition that we announced a few weeks ago. It’s a business with strong customer support, incredible data, critical use cases and a still nascent machine learning product proposition. There is much we can mobilise against here.

What determines a winner

The pace of AI adoption will be far from uniform. Across sectors, use cases, regions, different professional users, regulators and company sizes. We are paying close attention to how it’s developing in every sub niche that we cover. And vertical software companies will be critical in addressing the ‘last mile’ complexity for many. This is the messy reality of building AI products in the real world, with the endless edge cases, different regulations, data quality, and ways of working and getting that to paying customers from Toronto to Tallinn.

In mission critical vertical software, where we focus, the stakes are high too. Customers won’t keep using a product in payroll or accounting that works brilliantly most of the time, or a beta version that’s pretty close to working. As I started, this is where incumbents have a head start and the key is to have already mobilised.

This long-term perspective explains why public market sentiment is less critical to us than it might appear. As our Head of Research, David Toms, always reminds us, long term value creation in software over recent decades has come from earnings growth, not multiples. Capital invested for the next 5-10 years should be focused on the same tailwinds that have supported Hg for the last 25 years – workflow automation addressing the high cost of labour in western economies – but importantly now alongside an investor with the capabilities to understand, deploy and harvest the value of AI.

MEB