Investor note

The AI roadmap is not the moat

May 2026.6 min read

Audience: Software investors, partners, and operating partners weighing positions and underwriting moats in 2026.

The market claim

Every software company now has an AI roadmap, and most investor materials still treat that roadmap as the primary expression of competitive position. The roadmap shows assistant-style copilots, generative summarisation, smart automation, and an agent layer planned for the next two quarters. The market reads the slide as a moat, prices it as a moat, and weights it against peers as a moat. By Q1 2026, that reading had been priced through the public software basket; the iShares Expanded Tech-Software ETF was down more than 21 percent year to date and the sector erased roughly $2 trillion in market capitalisation (FinancialContent, The SaaSpocalypse). The repricing did not target companies with thin roadmaps. It targeted companies whose roadmaps could not show a defensible mechanism behind the features.

What investors are mispricing

The mispricing is structural.

Copilot isn’t a moat.

AI summarisation, AI-powered insights, and smart automation are now table stakes by 2026 across most software categories, and serve mainly as margin-improving features rather than competitive advantages (Designli, How to Build a SaaS Moat in 2026). The cost and time required to reach feature parity has collapsed because AI coding agents now generate the boilerplate, mimic the interface patterns, and automate the workflows that previously took a large engineering team months. A feature that competitors can ship in a quarter is not a moat. It is a maintenance item.

Three categories survive the repricing as structural moats in AI-native software: proprietary datasets that competitors cannot replicate, regulated and compliance infrastructure with multi-year switching cycles, and integration depth that makes the product a system of record expensive to remove (HarbourVest, The Software Industry’s Great Reset; Big Ideas DB, The SaaS Moat Crisis). These are not roadmap items. They are accumulated assets that take years to build and cannot be replicated. The investor reading that prices the roadmap and ignores the moat layer is solving for the wrong variable.

The company-level mechanic

Two operating signals show what the right reading looks like, and what mispricing looks like.

Mispriced cheap. Snowflake and Datadog remained resilient and in some periods posted gains against the broader Q1 2026 carnage, because each owns infrastructure-scale data and runtime gravity that an agent layer needs, not replaces (FinancialContent coverage). Cloudflare strengthened on the same logic because edge security and low-latency inference are non-negotiable for enterprise-grade agents. The roadmap layer in each case is unremarkable. The moat layer is structural.

Mispriced rich. A survey of 817 enterprise software builders found that 35 percent have already replaced at least one SaaS tool with a custom-built alternative, with workflow automation the leading category under replacement pressure (Big Ideas DB, The SaaS Moat Crisis). Vertical and horizontal workflow tools that publish an aggressive AI roadmap but sit on a generic data set and a generic integration footprint are the most exposed in the basket. The headline AI feature parity disguises a structurally weakening moat. That is AI roadmap theatre at the position level.

The mechanic is simple. AI shifts value from the application layer to the workflow layer underneath, and to the data layer underneath that. The application layer is where the roadmap sits. The workflow and data layers are where the moat sits. A company with a thick application and a thin workflow and data layer is mispriced rich. A company with a thin application and a thick workflow and data layer is mispriced cheap. The investor question is which one is being held.

Upside and risk

Upside. The investors winning this cycle are reading moat depth, not roadmap breadth. Thoma Bravo and Vista, two of the largest specialist software investors, are publicly arguing the right move is selective software buyouts of moat-thick companies at compressed multiples, while reassuring LPs that their existing portfolios are concentrated in the survivor categories (Bloomberg, Thoma Bravo and Vista; Yahoo Finance, Thoma Bravo and Vista Reassure Investors). Vista’s own portfolio is now reporting that the majority of its companies are not experiencing meaningful churn or losing clients to AI alternatives, consistent with the moat thesis. The trade is concentration in the structurally defensible quadrant, not basket exposure.

Downside. The reading-error case is owning a company that publishes the right roadmap and cannot defend the workflow underneath. The pricing model breaks first as customers force a renegotiation to consumption or outcome pricing, gross margin breaks second as inference cost is forced inside the contract envelope, and net revenue retention breaks third as agent deployment compresses the seat count. By the time the roadmap-led investor sees the underwriting error, the LBO model has lost capacity from three directions at once. ServiceNow’s transition to 50 percent non-seat new business and GitHub Copilot’s move to pure token billing on 1 June 2026 are the early operating signals that this sequence has started in the listed market (Bessemer, AI Pricing Playbook).

Watch this, not that

Watch proprietary data accumulation, regulatory and compliance integration, and the operational depth of customer integrations. These are the moats that survive the repricing. Stop watching the AI feature count, the demo quality, and the announced agent roadmap. These are the moats that the cycle has already proved insufficient.

Three diligence anchors for any 2026 position in software:

A documented inventory of the proprietary data graph the product has accumulated and the customer cost of recreating it. The cost is the moat.
A switching-cost classification by ARR cohort: regulated, data-graphed, integration-deep, or none of the above. The “none of the above” cohort is the cohort being repriced.
A read of where customer engineering teams would build internal alternatives if instructed to. The list is the inverse of the moat.

Position size on moat depth. Discount the AI roadmap. The cycle has done it for you.

Sources

moat, investors, diligence, multiples

Back to writing