Hg’s Value Creation and AI experts took to the GenAI frontier - Silicon Valley - to connect with foundation model providers, VC firms and the developer ecosystem rapidly growing around GenAI.

Over 30 meetings - ranging from foundation model providers, VCs through to the leading GenAI start-ups – showed that the pace of technology investment and innovation is only accelerating as the GenAI ecosystem comes together.

Below are some key observations and learning from various actors in the front line we met along the way:

The model ecosystem is maturing fast

There is recognition that some individual brands have been hyped, and that missing links in this emerging ecosystem have led to some broken promises.  But less visible is that some important pieces of the ecosystem puzzle are coming together fast. 

Most obvious is the continued progress of foundation models: more powerful ones (Claude 3 Opus), small & efficient models for local deployment (Mistral 7b), and multi-modality (GPT 4 omni). The GenAI Ops tools to better manage this technology are also in production – such as measuring & managing reliability.

The ‘mind the app gap’ is being addressed by the sheer number of new entrants focused on specific use cases.  There is undoubtedly some way to go, but we can expect progress on what is available and possible with GenAI to accelerate materially.

The foundation model race is also still in full flight. GPT-5 and other next-gen models are expected to drive continued step-changes in performance and features this coming year - the expected ‘end game’ is for continued leading positions from the ‘big four’ (OpenAI, Anthropic, Google, Meta), with a long tail (+1M!) of specialised models. 

Completing the GenAI ecosystem: putting in place the scaffolding

These models are increasingly viewed as ‘established foundations’, or even a ‘commodity’ for specific use cases, through easy and now automated routing to the right ones on a case-by-case basis. We see firms skipping more freely between model providers, prioritizing a broader range of factors such as latency and cost.

The real focus is now as much on the ‘scaffolding’ around these models to tackle the underlying business use cases – in particular, shifting the conversation from AI productivity tools to AI agents delivering complete tasks and workflows. This ‘agentic’ pattern appears to have potential to drive a paradigm shift towards new ways of interacting with software, and powerful early examples are emerging.

The looming battle to capture AI’s value

A battle to capture incremental value is well underway. The expectation is that the ‘AI layer’ will be the main beneficiary - and the driver behind much VC and start-up activity, with the hope to build on top of “soon to be commoditized” workflow and ‘system of record’ software providers.

Some AI-first entrants have started building the system of record themselves, and are considering offering these as ‘loss leader’ to drive adoption of more value-accretive AI features.

But these entrants are facing the familiar challenges of Go-To-Market and the Professional Services work to get adoption and drive value from their complex solutions.  This muscle is a key advantage many of the existing, established software providers are hoping to leverage, alongside proprietary data, instead acquiring these AI companies to merge into their offering.

Note that Sam Altman continues to hint that OpenAI can push up the value chain into these middleware and application layers, which would be bucking the consensus view and also where other foundation model providers are focusing.

Getting ready for the platform shift

Some of the largest incumbents are attracting noise due to their perceived failed starts. “The innovators dilemma” is a common diagnosis. However, this is still to be played out – for example, Oracle is now attracting keen interest, integrating more deeply with the Cohere team to build GenAI features into their products.

Of course these rapidly emerging ‘AI applications’ are worth tracking not just as potential future ‘software’ providers and competitors: there are many promising tech solutions emerging that suggest material operational efficiency upside to existing, including Hg, tech companies.

Above all else the Silicon Valley community is talking about this being a true ‘platform shift’ in progress, with ‘electricity in the air like nothing before’. 

With the AI ecosystem puzzle pieces coming together, the pace of innovation will only accelerate.  Our strategy is not to put too much weight on early disappointments or specific brands, but instead continue to build our own internal AI foundations, continue to experiment, and challenge ourselves to evolve our own operating models and products to crystalize impact.  The potential prize is a materially larger 'Target Addressable Market’ to grow into and a more efficient software firm of the future.