Let the AI come to you

Somewhere between downloading my fourth AI tool of the week and realising I hadn’t done any growth initiatives for Igloo all week, I had a thought: this is not productivity. This is just a different kind of busy.

The quiet tax of a technology craze isn’t the subscriptions, though those add up faster than anyone admits. It’s the attention. Every model drop, every connector launch, every workflow overhaul arrives with an implicit demand on your time, and unlike a bad ad campaign, it doesn’t feel like a mistake while it’s happening. It feels like staying competitive.

We’ve been here before, I think? When e-commerce exploded between 2019 and 2021, every brand with a Shopify store suddenly needed a tech stack that would make a Series B start-up blush. Then interest rates moved, DTC darlings started quietly returning investor capital, and only the tools that were genuinely useful survived or were adopted.

I don’t think it’s naive to assume the same shakeout is coming for AI, especially as we prepare for a wave of IPOs, this time for platforms burning through exponentially more capital than any DTC brand that went public, at market caps that would make even SoftBank do a double take. The question is whether the smartest move is actually to sit tight, find a better balance between growth and operations, and let the best options rise to the surface. Because if history is any guide, the winners won’t be the teams that adopted the most tools. They’ll be the ones that were still talking to their clients and building retention (maybe).

The triggers look familiar

The numbers behind AI adoption tell a story that should feel recognisable. Enterprise spending on generative AI hit $37 billion in 2025, up from $11.5 billion in 2024, a 3.2x year-on-year increase. For context, that trajectory mirrors almost exactly what happened to DTC and retail tech investment between 2020 and 2021, right before the correction. The difference is the scale. When e-commerce fell, it took Klarna’s valuation down 85% and wiped billions from the sector (taking brands with it). The AI equivalent, when it comes, could carry considerably more zeros (maybe a good time to join the church).

The more telling signal is what’s happening inside organisations right now. Call this lazy journalism if you will, but I’ve leaned on third parties for all this data, so forgive me. In 2025, only 31% of AI use cases studied reached full production, and expectations around cost reduction and productivity are broadly underdelivering. Meanwhile, a 2025 MIT study found that 95% of organisations deploying generative AI are currently seeing little to no return on investment. That gap between deployment and demonstrable value is almost identical to what preceded the 2022 SaaS cull. Procurement teams were signing off on tools they didn’t fully understand, for problems they hadn’t clearly defined. The rationalisation that followed was inevitable, and it was uncomfortable.

There is also a consolidation signal worth watching. By the end of 2023, OpenAI commanded 50% of the enterprise LLM market. By mid-2025 that had fallen to 25%, with enterprise dollars consolidating around fewer, higher-performing models. That kind of market compression doesn’t happen in a healthy, expanding landscape. It happens when buyers start making real decisions rather than exploratory ones. The experimentation phase is quietly ending. What follows it usually isn’t pretty for the tools that couldn’t justify themselves.

The cream always rises

It did in 2022, and it will again…probably. When the DTC market corrected, the tools that survived weren’t necessarily the best, or most funded (though this definitely helped with runway). They were the ones that had become genuinely difficult to remove, after spending months closing brands. Triple Whale kept improving its attribution modelling quietly and pushing the “better than Meta’s attribution” narrative while competitors fell a few steps behind developing features no one adopted. Klaviyo stayed focused on email and SMS retention and was quickly adopted by a huge chunk of the e-commerce market. Recharge built deep enough into Shopify’s infrastructure that removing it would have broken more than it saved. All of these tools acquired clients and made it really hard for a founder to feel obligated to remove them, even if the business started struggling.

The same pattern is already forming in AI. Code generation has emerged as AI’s first genuine killer use case, with 50% of developers now using AI coding tools daily, and the category growing from $550 million to $4 billion in a single year. That is a real productivity gain with a measurable output, not a dashboard metric. The tools winning that category, Codex, Claude Code, GitHub Copilot, are winning because developers can point to something tangible at the end of the day. That is the bar every other AI tool will eventually be held to. Not how well it presents in a demo, but whether someone would actually notice if it disappeared.

The “deploy first, strategise later” approach that characterised the initial AI rush is ending. What replaces it tends to be a much shorter list of tools, selected for specific, high-value use cases with clear stickiness from day one (granted, most are still burning cash, but at least it gives them grounds to re-raise). For e-commerce operators and marketing teams, that list probably includes the model you use for analysis, the connector that saves you three hours a week, and whatever your developers rely on to ship faster. Everything else is a subscription waiting to be cancelled. The brands and agencies that work that out before the downturn will be in a very different position to those that work it out during one.

Sources: Menlo Ventures State of Generative AI in the Enterprise 2025, ISG State of Enterprise AI Adoption Report 2025, MIT AI ROI Study 2025, Menlo Ventures Mid-Year LLM Market Update 2025