Cognition founder Scott Wu speaking to Harry Stebbings in a recent episode of the 20VC podcast

Despite frothy valuations and mixed receptions to recent model launches, to us Scott Wu here articulates why steady model progress remains our ‘longer-term base case’ when shaping B2B SaaS strategy for the coming years.

It’s easy to lose perspective just how far we’ve come: in the last three years LLMs evolved from amazing the world with Haikus, to models approaching human-level reasoning and demonstrating ‘mid-level’ autonomy in key verticals. Take Scott Wu’s Cognition here: their Devin agents are reliably delivering 2-10x leverage in our Hg software portfolio through autonomous coding.

There’s no consensus on what scaling compute, more data, or improved model techniques will contribute, with as many pointing to inflection points (‘AI lift off’) as there are those predicting plateaus. Our view: continued, steady progress should remain the ‘base case’. Domain-specific post-training is a vector we’re focused on in the shorter-term, and we’re watching carefully for technique break-throughs (especially in RL).

So what might white-collar work look like if the rate of progress from the last 3 years sustains for the next 3?​

As hopefully some of us get to take a step back over the summer break, we want to explore what such a scenario might imply for B2B SaaS. And even this base case already implies a significant step change by ‘28.

Software engineering agents give us a good look into the future of how white-collar work might get done in 2028: a small number of senior professionals and domain experts orchestrating (i.e. ‘managing’) a web of AI agents, each executing tasks end-to-end – tasks which previously required fully human teams.

In software, this means architects and product managers will define the intent, while AI agents implement, test, and optimise the code. Already today, we have an Hg portfolio business running with small squads of 1 product manager and 2 engineers - each ‘leveraged’ by up to 12 Devin AI agents. Scaling strategies shifts away from coding capacity, to product and operating vision.

White collar work in 2028: orchestrating agents

We should consider AI in software engineering to be a step ahead: it’s had the research focus, and closed-loops lend themselves particularly well to post-training improvements. But with research seeking to replicate those gains in other domains, it offers an interesting read across to other white-collar verticals:

For lawyers this might mean AI agents will handle document review, contract drafting and case research. A litigation team becomes one lead counsel orchestrating 10-15 specialised legal agents - trained on millions of sector specific precedents, regulations and case outcomes. The model shifts from billable hours to value delivered to their client. Leading global law firm A&O Sherman is already implementing their first set of ‘law products’ powered by AI.

For accountants, will AI agents execute audits, prepare tax filings, and generate regulatory reports? One controller could manage 8-12 accounting agents that reconcile transactions, identify anomalies, and produce real-time financial analysis. Value goes beyond compliance checking, to financial insight.

A business will rely on senior experts to lead functions (Finance, HR, Support), but with much more of the execution happening agentically, with humans providing oversight, checking edge cases, and managing the web of AI agents.

Orchestrating agents: becoming the system of action

This fundamental transformation will shift software from “running the business" (contract change management, billing) to software that "does the work itself" (negotiating legal contracts, completing audits). The more actions you own and perform on your own data, the more value you add for your customers.

The prize is for those who grow into and become that platform where AI agents execute work, make decisions and trigger downstream processes.

Tidemark coined this as the “system of action” that will orchestrate the collaboration between humans and AI-agents, capturing both workflow control and the data exhaust from millions of automated decisions.

Today, B2B SaaS only owns “systems of record”, the human workflows that they enable are the “system of action” for their customers. The goal is clear - to offer agentic solutions that move up into this “system of action.” This pits incumbents - with deep domain knowledge and established customer relationships, but still learning how to build agentic AI products - against emerging AI-native startups who are borrowing or buying domain expertise, and establishing customer relationships fast.

The winner will be those who capture not only the data, but also where the work gets done - the agentic AI cockpit, acting for the human pilots.

The race to become the system of action has started - yet many incumbents don’t even know they’re competing

AI research labs seeking out monetisation opportunities are actively pushing into some select vertical applications – especially coding, and some key consumer applications. But for most niche B2B verticals they’ll be the AI partner to help deliver the system of action: navigating that ‘fractal complexity' remains the right to win for vertical specialists.

AI-native start-ups have been prosecuting verticals from ‘day 1’ (now >70k AI startups). What we might initially dismiss as ‘simple wrappers’ or ‘point solutions’ are gaining traction at the thin edge of the wedge, and fueled by Product-Led Growth and VC CapEx they will quickly look to land and expand to capture more value.

They also view AI expertise as their edge that matters in the race over the incumbents, partnering with end customers or also borrowing in experts to overcome domain expertise gaps.

Many B2B SaaS incumbents remain oblivious, even cooperating naively with AI start-ups who view them as future 'tool calls' rather than partners.

But we’re also seeing ‘enlightened’ incumbents bundling in AI, leaning on their customer and domain head start to catch up on core AI skills.

7 essentials to win the race

The starting pistol has fired silently, the race is on.

B2B Tech incumbents typically already enable today’s “system of action”, often already embedded with customers, trusted to help deliver key processes. The challenge is to bring in AI value-add and agents to maintain that position, and use the new AI tools and data flywheels that can be built to capture materially more value.

They still hold the advantage, and also longer-term strategic value creation opportunity, but seizing this opportunity for the future, starts with critical actions today:

Shift board conversations today to drive the product innovation imperative we now face. Backwards-looking board reviews of financial performance will miss the race that’s unfolding in front of them.

Map out your customers’ key pain points and workloads and form a view of the likely role agents will play with your customers and in your ecosystem. Merely tuning and extending existing propositions will miss the step-change.

Develop an AI product vision and roadmap; also identify quicker wins to ‘mark your territory’. Urgency matters.

Invest in your AI product muscle to develop the talent and processes to drive the product innovation acceleration. Confront the new organizational bottlenecks early.

Invest in high-quality AI engineering talent to deliver AI product builds. Recognize the new skills and ways of working needed in R&D.

Understand your existing SaaS stack’s ‘AI readiness’ to integrate with agentic layers via API and handle the new type of workload. New pressure points will emerge.

Mobilise an operational AI transformation to not just capture margin but also bring the agility and pace to compete. AI products require an AI-first org to support it.

Think longer-term, despite short-term uncertainty

With such uncertainty and a wide fan of outcomes it is tempting to wait on the sidelines for the fog to lift. But we’d challenge that view strongly: even if the exact shape of the AI technology progression remains unclear, that B2B SaaS can play a more agentic and higher value-add role for customers is clear, as are the key steps needed today to capitalize on the emerging opportunities. This is the change we're accelerating across Hg's portfolio.

Ready to help us drive this next platform shift?

We're investing heavily in our GenAI flywheel through exceptional talent, our Hg product incubator, and pioneering AI partnerships. Join us on this transformative journey - get in touch: AI@hgcapital.com!