In an article published by Salesforce, it is argued that many organizations building with artificial intelligence drop an agent into a single step of a workflow, while the steps before and after it continue to move at the ordinary human organizational pace. The article compares this situation to a race car with a spectacular engine that finds itself stuck behind a tractor on a country highway, explaining that while the agent finishes its work in seconds, the work then waits for a review, a handoff, or a change request. According to the article, this is why 95% of companies are not deriving value from artificial intelligence, as value gets trapped at every human-scale step the agent is forced to wait on.
Workflow Bottlenecks and Human Pace
The article explains that one can think of reasoning across every client health conversation that has ever happened, searching through 27 years of Salesforce data, and receiving an answer—work that no human could perform. However, if one step is made agentic, the step following it must become agentic as well; if it does not, the process remains stuck at human scale, behind the tractor, and that is where return on investment stalls out.
According to the article, the fastest way to become an agentic organization is to create an open, secure, and governed environment where every employee has the opportunity to build. At Salesforce, 74% of employees have used or interacted with an agentic coding tool, such as Claude Code or Codex, at least once. Usage among non-technical employees reaches up to 66%, a figure representing more than triple the usage in the first quarter. Most of them do not know how to write code, but they are willing to experiment with explaining to a machine what they need.
The Risks and the Role of the IT Department
Alongside this, the article emphasizes that without the right methodology and infrastructure in place, a situation where everyone is building everything everywhere could turn into a security nightmare. An agent lacking appropriate guardrails could make mistakes, do so with complete confidence, and act on that error at machine speed.
Furthermore, the IT department is still required to own responsibility for trust, compliance, and integrity. IT must continue to own the systems of record and data semantics, such as the definition of what a customer is or what revenue means. The article notes that all of these can be exposed through APIs and MCPs, because current building is not intended only for humans, but primarily for agents that people will use outside of IT control, building, or ownership.
The Rapid Innovation Squads (RIS) Methodology
To address these challenges, Salesforce developed a model known as Rapid Innovation Squads (or RIS for short). In early 2026, the author and their team realized that agentic transformation extends far beyond tools alone, raising questions about whether product people, business stakeholders, and engineers could work together as a single small team co-building a solution, without a heavy work program and without a pre-baked product requirements document. The goal was to examine how much faster humans and agents could move together, given the right structure, guardrails, and tools, while maintaining security, and what redesigning teams, processes, and systems for agentic scale would look like.
According to the article, the structure of a RIS consists of an opinionated and empowered business stakeholder who owns the problem and holds clear views on how to solve it. Alongside them join agentic builders—whether one, two, or five individuals—keeping the team as small as possible, while the problem is reduced to its smallest useful form.
The Daily Routine of RIS Squads
The article details the daily routine of a RIS squad:
- Joining the morning stand-up: The squad convenes for a daily stand-up meeting that is recorded and transcribed.
- Transparency by default: The transcript is posted to a public Slack channel so that the rest of the organization and all agents can learn from the squad's work. The iteration process takes place there as well: employees comment, suggest changes, and build on their peers' work within the thread (for truly sensitive projects, the channel can be private).
- Handing off to agents: The transcript is fed into an agentic coding tool. The team steps away, and the agents begin working on what was discussed during the meeting.
- Repeating after lunch: Every team member functions as a builder, and everyone arrives with ideas and prototypes.
- Repeating at the end of the day: The agents continue building, testing, and iterating overnight, ensuring the team opens each morning with working progress to review rather than an awaiting backlog.
According to the article, the payoff is speed: partnering with an opinionated stakeholder leads to decision-making that is about 10 times faster, and combining this with agentic coding tools produces outcomes about 30 times faster. The article suggests viewing speed as a proxy metric for value delivery when executed correctly.
Projects and Implementation Examples at Salesforce
The article presents several examples of the model's implementation within Salesforce:
- Sales Signals app in Slack: Small and medium-sized business (SMB) sellers faced context switching and lost time moving between multiple tools. Buying signals—such as website visits, usage spikes, and contract renewal windows—were scattered across disparate systems, and the team sought a persistent workspace inside Slack to consolidate all signals into a clear daily work list. Instead of handing off a 12-month project plan to engineering, the problem was assigned to a three-person RIS squad operating on a 24-hour cycle of planning, building, swarming, reviewing, and shipping. The squad delivered the Sales Signals app within days, whereas a traditional project plan would have stretched across months. The app has already generated 28,000 business opportunities.
- Slack Skill for solution engineers: Solution engineers at Salesforce used to spend two to three hours per week logging their activity through an error-prone process. In less than four weeks, a three-person squad (one business leader and two agentic builders) shipped a new Slack Skill that scans the engineer's calendar, Slack data, and Salesforce systems to pre-fill a timesheet for the engineer's approval. The sprint produced the tool 70% to 80% faster than a traditional Agile process, adoption stands at 80%, and 23,000 weekly activities are now automatically logged.
- Case routing in the legal department: The legal department routed and triaged approximately 39,000 cases annually, with routing each case to the appropriate attorney taking around 20 minutes, alongside occasional duplicate routings that caused attorneys to waste time on cases someone else was already addressing. A small squad combined deterministic rules with Agentforce to route cases almost instantaneously, saving $350,000 per year—a 70x return on investment that paid for itself within a month. Furthermore, the agents de-duplicate active inquiries and remember full historical handling to refine routing accuracy, yielding routing that is nearly three times better at selecting the correct attorney on the first try.
Companywide Expansion and Workflow Recommendations
The article states that in December, the model expands across the company through an event dubbed RISFest—an open call to every Salesforce department to submit a real business problem for a RIS squad to build a solution for in days rather than quarters. The initiative is defined as an "anti-hackathon" aimed at shipping solutions directly.
The article outlines practical recommendations for managers and organizations:
- Do not settle for automating an isolated task; redesign the entire workflow so that every step operates at agentic scale.
- Start with a problem and an opinionated stakeholder who owns it.
- Keep the team small and the problem even smaller.
- Provide creators, whether technical or non-technical, with safe building blocks, while keeping the IT department responsible for trust, compliance, and data infrastructure.
- Work openly in public so that every squad learns from the one before it.