Our recent research-driven investment trip reinforced how AI is delivering real efficiency and productivity gains — but outcomes are uneven. The strongest operators are winning not with model access, but on implementation: data quality, workflow design, and the judgment to know when to trust an output and when to reject it. That unevenness makes selectivity key for investors.
In early summer a group of our analysts and portfolio managers spent several days in San Francisco as part of our ongoing research into artificial intelligence (AI) and its impact on investment opportunities. The trip reinforced a simple point: AI is now producing real – but highly uneven – business outcomes.
We met several companies across software, healthcare, industrial and consumer sectors with a consistent message emerging. The experimental phase is giving way to measurable productivity gains in selected workflows, yet broad-based transformation remains slower, more difficult and more company-specific than market enthusiasm suggests. For investors, this distinction matters and the next phase of AI adoption is likely to reinforce the benefits of selectivity over broad based exposure.
AI driving operational efficiency
The clearest gains from AI deployment are arising where tasks are repetitive, data-rich and easy to evaluate. At one consumer platform we met, an AI-enabled support tool is already resolving around 40% of customer queries without human involvement. The implications for servicing costs are obvious. In life sciences, generative AI tools for designing molecules are expanding the number of drugs viable for early-stage research, increasing the odds of successful development over time. Elsewhere, AI is helping non-software specialists build simple internal tools themselves, compressing tasks that previously required external vendors or specialist teams. These are not abstract possibilities – they are operational improvements with clear economic potential.
Key takeaway:
- The next wave of AI winners we believe will be companies applying tools to discrete, measurable workflows where productivity gains can be proven, sustained and scaled.
AI coding gains traction
Software development is an area where we see rapid progress. Several management teams described a sharp inflection in developer productivity over the past year as coding tools improved and adoption broadened. In some cases, a meaningful portion of new code is now being written autonomously or generated with heavy AI assistance. One company explained how ~10% of production code is written by AI while another highlighted that around 75% of new code was AI produced – an increase from 50% 12 months previously. Corporate functions are moving more slowly, but the direction of travel is the same. Finance automation was also a recurring theme: routine reporting and weekly performance summaries that once took days can now be completed in hours. The savings are still narrow in scope, but the signals are increasingly credible.
Key takeaway:
- Evidence suggests AI can drive meaningful productivity gains in coding and back-office functions. Companies that can scale and execute are well positioned to translate automation into improved margins.
AI benefits – from theory to reality
Almost no company claimed to have a fully mature framework for measuring AI returns. That remains a work in progress. But the discussion has moved on from whether the investment is justified in principle. A year ago, many projects were funded on conviction alone. Today, management teams are seeing enough evidence to believe returns are positive, even if they cannot yet quantify them precisely. The framing has evolved from “we believe this will work” to “we know it works – we just don’t know how much yet.”
That is an important shift. AI is no longer a pure leap of faith. Equally, it is not a universal profit lever. The payoff depends on where it is deployed, how well it is integrated and whether the organization can adapt around it.
That leads to a more important conclusion for investors. Competitive advantage is likely to depend less on access to models and more on implementation capability. The strongest operators were not fixated solely on model performance. They were focused on data quality, workflow design, evaluation metrics and human oversight. In research-heavy industries, this point was especially clear: asking better questions, setting the right success criteria and knowing when to reject an output are becoming core skills. AI changes the role of expertise, but it does not remove it. In many cases, it makes judgment more valuable, not less.
Key takeaway:
- There is increasing visibility into returns from AI investment, but the benefits are not universal. Early movers that successfully execute and integrate AI are likely to have an advantage.
SaaS’s structural disruption
The trip also highlighted that AI spending is forcing tougher choices elsewhere. Companies are funding this investment through slower hiring, tighter cost control and closer scrutiny of software budgets. That is particularly relevant in enterprise software. The distinction many management teams made was between systems that hold critical data and point solutions that sit around them. Platforms that act as a source of truth appear relatively well protected because the data resides there. Narrower applications face a more uncertain future, especially where similar outcomes can be achieved through broader AI-enabled workflows layered on top of core systems. This raises the prospect of growing pressure on pricing, product breadth and customer retention across parts of the software stack.
Key takeaway:
Platforms with embedded data are better positioned amid an AI-driven reset for software providers. Point solutions appear more vulnerable.
Cost considerations drive model choices
Another useful insight was the degree of pragmatism around model selection. Leading providers retain clear advantages, but enterprises are not simply defaulting to the most prominent names. They are comparing models on cost, speed and task-specific performance, and are willing to choose lower-cost alternatives where the quality gap is small. One company chose a lesser-known model for review synthesis as the quality differences did not justify the premium. That matters because it suggests the market may not settle around a single dominant pricing structure. Instead, value may accrue across a broader set of providers and tools, depending on the economics of the use case. For investors, that makes the landscape more nuanced than a simple scale-wins narrative.
Key takeaway:
Demand is shifting toward best-fit AI models, as users prioritize utility and cost over frontier capabilities. A broadening field of AI providers reinforces the importance of stock selection.
The bottom line
Our trip and dialogue with companies highlighted how AI is becoming economically meaningful. Its benefits, however, are proving narrower, more operational and company-specific than the broader narrative suggests. While some firms will be constrained by data quality, weak workflows, tighter budgets and limited organizational readiness, the winners will be those with the data, discipline and execution capability to turn adoption into durable returns. Against this backdrop, we see opportunities for selective investment.
