Salesforce says AI stalls when it speeds up one step and the rest of the process stays slow. Its fix is tiny teams that build with AI tools.
Salesforce says many companies get little from AI because they speed up one step while everything around it stays slow. Its answer is small teams, called Rapid Innovation Squads, that build tools in days with AI coding helpers. The company says this delivers results about 30 times faster. These figures come from Salesforce itself, not from an outside review.
Salesforce compares it to a race car stuck behind a tractor. An AI agent, which is software that does tasks on its own, may finish its step in seconds. Then the work waits for a review, a handoff or a change request. The company links this to the claim that 95% of companies are not getting value from AI. The value gets trapped at each step where a person has to catch up.
Each squad starts with a business owner who has a real problem and strong views on how to fix it. Then one to five builders join, and the team stays as small as possible. The problem is shrunk to its smallest useful form. The daily rhythm looks like this:
Salesforce says the business owner brings decisions about 10 times faster. Adding AI coding tools brings outcomes about 30 times faster.
One three-person squad built a Slack app called Sales Signals in days. It gathers buying signals, such as website visits and renewal windows, into a daily work list for small-business sellers. Salesforce says it has generated 28,000 opportunities. A traditional plan, the company says, would have taken months.
Another squad, with one business leader and two builders, built a timesheet helper in under four weeks. Solution engineers used to spend two to three hours a week logging activity. The tool reads their calendar, Slack and Salesforce data and pre-fills a timesheet for approval. Salesforce reports 80% adoption and 23,000 activities logged weekly.
Salesforce says 74% of its employees have tried an agentic coding tool such as Claude Code or Codex. Use among nontechnical staff is up to 66%, more than triple the first quarter. But it warns that an agent can be confidently wrong and act on that error at machine speed. IT still owns trust, compliance and the systems of record, including what terms like customer or revenue mean.
The lesson is not tied to Salesforce's size. Before adding AI to a task, look at what happens before and after it. If the next step still waits on a meeting or a sign-off, the gain may vanish. Start with one owner, a very small problem and a very small team.
Adding AI to one step helps little if the steps around it stay slow, so look at the whole process and start with a small team.
Otto, a new customer-tracking app for field sales teams, lets reps speak their notes instead of typing them. Price and launch details are not yet known.
Read more: Otto launches a sales app you talk to instead of typingWorld models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasks
