ISSUE 04 / OCT 5, 2026
This week in enterprise AI: what happened, why it matters, what I'd do.
Stop counting agents.
Kyndryl, Kaiser Permanente, Ally and a $20M lesson from healthcare: what lasts is how you run AI, not how much of it you count.
|
■ FROM THE PEOPLE DOING IT “Year one, we were all about counting agents, and that wasn't telling me the true story.” Kim Basile, CIO, Kyndryl |
|
PATTERN WATCH Kaiser bargained AI with its unions before rolling anything out, and Ally handed AI to everyone to take the fear out of it. Both started with people. The healthcare story shows what happens when the bet is on the model alone. |
|
|
A year into agents, the early adopters have stopped counting them. What they report now is what moved. Two of this week's stories are about people before technology, and one is a reminder that the model itself is the easiest thing to copy.
— Manny
|
| THIS WEEK'S THREE STORIES |
|
|
■ CASE STUDY · CROSS-INDUSTRY
Fewer agents, mapped to real work.
What happened Kyndryl CIO Kim Basile says year one was all about counting agents, which didn't tell the real story; Kyndryl now maps business processes instead. Salesforce cut from hundreds of internal agents to a few with clear benefits. Mining company Freeport-McMoRan ran about $500 million of invoices for one product line through an agent over six months. Adobe judges its help-desk agent, which serves 30,000+ employees, on accuracy, satisfaction and time returned, as well as ticket speed.
Why it matters Counting agents is the activity scoreboard. The early adopters have moved to the outcome scoreboard, and some cut their agent count to get there.
WHAT I’D DO Replace any “number of agents” line in your AI report with one process outcome, such as days to close or invoices processed. |
TO MAKE THIS WORK First: Give each agent a process and an owner when it goes live. An agent with neither should be the first one you cut. As you do it: Agree with finance how each outcome is counted before you report it. Then: Review those outcomes on a fixed schedule, and retire agents that don't move them. |
Read the source →Fortune, Sep 29, 2026
BY ROLE Executive sponsor: Ask for outcomes, not activity, in every AI update you receive. Transformation lead: Map business processes before adding agents, as Kyndryl did. Finance and risk: Measure AI by business value, such as closing the books in three days instead of five. | BEHIND THE NUMBERS A Salesforce legal-team workflow raised correct first-time routing of new cases from 12% to 34%, saving over $350,000 a year. Kyndryl now measures AI by business value, such as closing the books in three days instead of five. Deloitte: 74% of leaders expect nearly half their processes redesigned around agents within four years; 16% feel prepared. | HOW SOLID IS THIS Company-reported to Fortune. |
|
|
|
■ CASE STUDY · HEALTHCARE
Kaiser Permanente gave 62,000 unionised workers a say in every AI decision.
What happened Kaiser Permanente and the Alliance of Health Care Unions, which represents 62,000 nurses, technicians and other staff, agreed to give workers input at every stage of AI decisions, from choosing problems to training staff. A task force of 10 senior leaders, five from the unions and five from Kaiser, has direct authority over technology investment. More than 3,500 front-line teams will look for places AI could help, and peer advisers will learn each tool first, then coach colleagues.
Why it matters Trust can't be mandated. Kaiser had the conversation first and brought in the technology second, the reverse of most rollouts, and used people workers already trust to carry it.
WHAT I’D DO Before your next AI rollout, sit down with the people whose work it changes and agree what moves to AI and what stays theirs. |
TO MAKE THIS WORK First: Put worker representatives on the group that approves AI spending before the first project is chosen. As you do it: Agree role by role what moves to AI and what stays, and write it down. Then: Ask front-line staff where the new process breaks. They see problems the plan misses. |
Read the source →MIT Sloan, Sep 28, 2026
BY ROLE Executive sponsor: Give front-line representatives real decision rights on AI investment, not just a briefing. People managers: Bring managers and front-line staff into design, not just training. HR and L&D: Build peer advisers who learn each tool first and coach colleagues. Kaiser reused the model from its medical records rollout. | BEHIND THE NUMBERS Before formal bargaining, 300 union leaders and 80 Kaiser leaders met to set expectations. Union reps asked for an inventory of the AI tools already in use, including in hiring and performance monitoring. The peer-adviser model comes from Kaiser's electronic medical records rollout 20 years ago. Kaiser is making AI pilots visible across the organization to cut duplication. | HOW SOLID IS THIS Reported by MIT Sloan from the parties' own accounts. No results data yet. |
|
|
|
■ CASE STUDY · BANKING
Ally gave AI to every employee, inside its own walls.
What happened Ally Financial, a digital-only bank, made AI available to every employee to take the fear out of it and surface practical uses. All of it runs inside Ally's own cloud, so customer data is never used to train outside models. CISO Spencer Cremers splits AI security three ways: using AI to defend at a scale people can't, defending against attackers who use AI, and keeping Ally's own AI use safe. Ally also runs company-wide crisis drills with the CEO; the latest lasted three days.
Why it matters Giving everyone AI reduces fear and surfaces real uses, but only once the boundaries exist. Ally decided where its data lives before it handed out the tools.
WHAT I’D DO List where your AI tools send company data today, and which could train on it. Fix that before you widen access. |
TO MAKE THIS WORK First: Find out which AI tools staff already use on their own, and what company data goes into them. As you do it: Set the limits in the tools themselves: where data is stored, how long it's kept, and who can connect what. Then: Give access to one group first and see what they do with it before everyone gets it. |
Read the source →Technovation with Peter High, "Security by Design: How Ally Financial's CISO Builds Resiliency Into the Cloud", Sep 28, 2026
BY ROLE Technology and data: Keep AI inside your own cloud tenancy wherever customer data is involved. Finance and risk: Add AI-driven attacks to your next crisis simulation, with the CEO in the room. Executive sponsor: Once the guardrails exist, give AI to everyone. Holding it back mostly feeds fear. | BEHIND THE NUMBERS Ally puts cybersecurity and technology risk under one CISO, so controls are designed in rather than bolted on. It is adding AI security tools that didn't exist six months ago, accepting near-term complexity it may need to rationalize later. | HOW SOLID IS THIS A CISO describing his own bank's approach on a podcast. No figures given. |
|
| THE PROOF IS IN THE PUDDING |
|
33 of 114 33 of 114 companies showed a number for what AI did for their organization: 29%. One-week average: 29%. The rest said they use AI but showed no result: no cost cut, no hours saved, no error rate. One that did: When 7-Eleven introduced AI agents to automate 95% of its store hiring, it reduced time to fill by 70% and saved store managers more than 2 million hours annually (Harvard Business Review) |
|
A $20M custom AI build the next model release made easy to copy. A venture-backed precision-medicine company spent about $20 million over three years building a custom AI product. When GPT-5 launched in August 2025, any competitor could replicate essential elements of it at a fraction of the cost.
Harvard Business Review, Sep 28, 2026 Act: California's “No Robo Bosses” law: AI can recommend a firing, a person must decide. SB 947 takes effect July 1, 2027. Employers may not rely only on automated systems to discipline or fire workers, and tools that predict employee behavior from personal data are banned. Fines are $500 per violation; wrongful-termination suits may be the bigger risk.
CIO.com, Oct 2, 2026 Act: 54% of organizations have no defined limit on AI agent access, or simply reuse a person's access. In a Gartner survey of 297 security leaders, they either have no defined approach or simply reuse the access of the person involved.
Gartner, Sep 30, 2026 |
|
What are you being asked to decide about AI, or where is a change stuck? Send me your question. I’ll answer the best ones in future issues, and I never use your name. |
|
■ TRANSNOVA.AI
Where is your team stuck?
AI isn’t the transformation. Redesigning the work is. Transnova.ai, my firm, finds the workflow that keeps slowing your team down, redesigns it, and measures the result.
|
|
Invite me to speak · Connect on LinkedIn
|
|
