Salesforce targets December for companywide rollout of small AI-building teams

Salesforce plans to open its small-team AI development model to every business function in December 2026. Through an initiative called RISFest, employees will submit business problems for Rapid Innovation Squads to tackle. The goal, according to enterprise and AI technology president Joe Inzerillo, is to ship working solutions in days rather than quarters.

Fewer handoffs, not just faster code

In a November 4 post, Inzerillo argues that speeding up one task accomplishes little if the next step still waits for a review, handoff or change request. His proposed fix is organizational: bring the person who owns the business problem into the same small team as the people using AI tools to build its solution.

The squad starts with an empowered stakeholder who has strong views about the problem, then adds as few builders as possible. Inzerillo frames the change as a way to combine product, business and engineering work without a formal program or prewritten product requirements document. The problem itself is reduced to its smallest useful form.

Rather than handing a finished specification to engineering, the team repeatedly turns its conversations into instructions for AI coding tools. Salesforce describes a daily cycle:

• Record and transcribe the morning stand-up, then post the transcript in an internally public Slack channel. Sensitive projects can use private channels.
• Feed the transcript into an AI coding tool. Colleagues can suggest changes and build on the work in the Slack thread.
• Repeat after lunch and at day's end. Agents continue building and testing overnight, giving the team working progress to review the next morning.

This is the anti-hackathon: We’re straight up shipping solutions.

Joe Inzerillo, Salesforce president of enterprise and AI technology

The internal projects behind the rollout

Salesforce's internal examples put that approach against longer development plans. Small-business sellers needed a daily work list inside Slack because buying signals were scattered across systems. A three-person squad delivered the Sales Signals app in days, rather than following a 12-month engineering plan. Inzerillo says the app has generated 28,000 opportunities; that figure describes opportunities, not closed sales.

Another three-person team built a Slack tool for solution engineers in less than four weeks. It draws on calendars, Slack data and Salesforce systems to prepare activity records for engineers to confirm and approve. Previously, logging that work took two to three hours a week and was prone to mistakes.

Salesforce puts adoption at 80%, with 23,000 weekly activities logged automatically. It also says development was 70%–80% faster than a traditional Agile process. These are company-reported results; the post does not explain how that development-time baseline was calculated.

For legal work, a squad combined fixed rules with Agentforce to route cases and catch duplicates. Inzerillo says routing previously took about 20 minutes per case across roughly 39,000 cases annually. He credits the system with almost instantaneous routing and $350,000 in annual savings.

More builders, but not ownerless systems

The companywide push comes as Salesforce says 74% of employees have used or interacted with an AI coding tool at least once. Among nontechnical employees, the figure is 66%, more than triple the first-quarter level. Those measures capture exposure to the tools, rather than showing that every employee regularly builds software.

Inzerillo's governance boundary is explicit: IT must still own trust, integrity, compliance and systems of record. It also controls shared definitions, such as what counts as a customer or revenue. Business teams get room to build, but not authority to redefine the underlying data foundation.