In recent weeks, Anthropic has unveiled a series of functions such as code writing, collaboration, Excel and legal review. Anthropic’s moves have fed into talk of an AI-driven crisis in the SaaS industry, with a wave of claims that SaaS growth and sustainability are being shaken.

The impact is not limited to SaaS. Some analysis says Anthropic’s new product launches are also directly pressuring the vertical AI ecosystem. The view is that as foundation model (FM) developers such as Anthropic bolster specialised expertise across areas to expand platforms and boost revenue, companies that focus on vertical AI are coming under direct influence.

Nihar Bobba (nihar bobba), a partner at Better Tomorrow Ventures, also sees the moves by foundation model companies as a threat to vertical AI firms. In a recent post on social media platform X, he said vertical AI with a “last mile” that is not long could be especially at risk.

He said the “last mile” will be the decisive battleground, referring to the final stretch that links intelligence produced by AI to specific outcomes.

The last-mile stretch includes real-world responsibility and regulation, administrative procedures and integration with external systems. It is the stage of handling exceptional situations a model did not predict in advance and completing digital outputs into real-world results.

The length of the last mile differs by domain. Generating marketing copy or producing images is the shortest. These areas are transaction-type businesses that provide simple outputs. He said foundation models have already absorbed them or are likely to do so soon. Coding assistants, basic legal drafts and research copilots have a slightly longer last mile than marketing. They are useful, but their defensibility is weak.

Contract analysis, financial modelling and internal workflow automation fall into the middle range. “In this area too, foundation model companies are planting their flag quickly,” he said. “A vertical AI company that starts in this band will see its position narrow unless it moves to a longer last mile. The time to move is also getting shorter.”

By contrast, tax filing and audit responses, compliance accounting and licensed professional services have a long last mile. “Cross-border regulatory coordination, engagement with administrative agencies and linking to real systems are essential,” he said. “Areas like healthcare, where legal requirements are clear, also fall here. These areas are structurally defensible.”

Even among vertical AI targeting the same industry, the length of the last mile can vary by product. Among vertical AI, tools that help lawyers have a short last mile.

“In lawyer-assistance AI, responsibility for results remains with the lawyer,” Bobba said. “If AI-native legal services directly take on responsibility for results and liability for compensation, the situation changes. In tax software, it is vulnerable if it only provides calculations. If it takes on calculation, filing and audit responses, it enters a long last mile.”

Bobba stressed that where the last mile sits is determined by 3 questions.

He said the questions are: Who is responsible for the outcome? Who bears responsibility when a problem occurs? Who do regulators or auditors contact? If the answers to all 3 are the company, the last mile is long. If the answer is the customer, the risk of foundation models pushing in increases.

Taking on responsibility can look burdensome for service providers, but in vertical AI it is a strategic advantage. “If you own responsibility for the outcome, you can compete on trust,” he said. “The pricing structure also changes. Customers pay for certainty instead of software fees. That is pricing closer to insurance. Those barriers build over time.”