Discover
YPO Technology Network AI Brief
YPO Technology Network AI Brief
Author: Stephen Forte
Subscribed: 26Played: 171Subscribe
Share
© Copyright 2026 Stephen Forte
Description
AI moves fast. Your briefing should move faster. The YPO Technology Network AI Brief is a daily breakdown of the AI developments that actually matter to your business. No hype, no jargon, no filler — just what changed, what it costs you or saves you, and what to tell your team on Monday. Hosted by Stephen Forte for the leaders who don't have time to chase the news but can't afford to miss it.
202 Episodes
Reverse
A viral post said Google "mathematically proved" AI will destroy the economy. The Google DeepMind paper behind it is a model, not a forecast, and its real point is more useful: agents are getting smarter much faster than they're getting better at working together. Anthropic learned that when its AI shopkeeper and AI boss spent a night chatting about "eternal transcendence," and procedure is what helped. Stephen shows how he runs agents so every step leaves a receipt.How to coordinate agents todayRun work in phases. Each agent leaves a receipt you can open (a document, spreadsheet, GitHub commit or ledger entry), and the next agent starts from it.Check every handoff with a second agent or a simple rule. A daily audit flags anything filed under the wrong client and alerts a person. Put a person where a decision is made or something leaves the company.See inside: the audit keeps its own log, so once it flags a problem you can see exactly what happened, step by step.Swap the part: if every phase leaves the same receipt, the model behind it is replaceable. Stay model-agnostic.Moves for this quarterPick one recurring process, draw its handoffs and name the receipt each step leaves, stored in your systems.Write a one-page contract for each receipt. You own the contract; the model is replaceable.Run it attended first, read the receipts weekly, then let it run unattended.Measure the handoffs, not the seats: where work stalls, what gets sent back, which step broke.For the CEO. You still manage people, who now manage agents. Drop the five-year plan and get hands-on. If your business moves bytes, not atoms, a six-person billion-dollar competitor is a real threat.Sources: Organizing Intelligence; DeepMind essay; Anthropic Project Vend; Anthropic multi-agent research; Stanford Digital Economy Lab.The AI Brief from the YPO Technology Network is a daily executive briefing on the AI developments that matter to business leaders. Hosted by Stephen Forte.
Cursor's breach didn't start with a bug in the app. It started with an access key Cursor issued to a vendor, Fireworks. That key opened old coding-agent session logs, some still holding pasted credentials, which led into support tickets with more keys. On 29 October, fake Claude Code extensions sat in Cursor's marketplace for about 11 hours. Stephen's take: check your machines, never paste credentials into a chat, sign in instead of sharing keys where you can, cap every AI account, make vendor keys expire, and keep a record of every key you issue.CEO checklistCheck for the fake extension. Anyone using Cursor opens Extensions, finds Claude Code and sends IT the version. If it's 2.1.312 or 2.1.313 from 29 Oct, follow Cursor's security notice. Uninstalling alone may not be enough.Turn on Privacy Mode in Cursor. Rotate any keys Cursor emailed you about.Never paste credentials into AI chats, support tickets or team channels. Treat any key that was ever pasted as compromised and rotate it.Treat agent session logs as sensitive data: who reads them, how long they're kept, whether secrets are stripped.Where a tool supports Anthropic's linked billing (beta), sign in instead of sharing a Claude API key, and set a per-tool spend limit.Set a monthly spend limit and an email alert on every AI account. On OpenAI, turn on "Enforce a hard limit."Vendor keys: limit what each can do and reach, expire them in a day or a week, and reissue.Keep a register of every key, ideally in a secrets manager: who has it, what it reaches, when it expires.Sample prompt: fake extension check. "Without installing, changing or deleting anything, tell me which version of the Claude Code extension (Anthropic.claude-code) is installed in Cursor on this computer, and when it was installed or last updated. Tell me clearly if it's 2.1.312 or 2.1.313, or if it changed on 29 October 2026."Sample prompt: key audit. "Audit this computer for API keys, access tokens and other credentials. Never print, copy, send or change any secret value. Only list where each is stored, which service it's for and when it last changed. Flag any in chat logs, notes or documents. Don't delete or rotate anything."Watch-outs. Linked billing is beta, works only in supported tools and only for organizations billed by Anthropic directly, and your own retention settings don't apply to those requests. Spend limits aren't instant. Stolen AI keys are resold: Google found fraudulent charges above $100,000 in a day.Sources: Cursor update, 13 Nov; Anthropic linked billing; Claude spend limits; OpenAI spend limits; OpenCode breach; TechCrunch; Google on LLMjacking.The AI Brief from the YPO Technology Network is a daily executive briefing on the AI developments that matter to business leaders. Hosted by Stephen Forte.
Microsoft Ignite's biggest news was plumbing: who each agent is, who it reports to, what it can touch and what it costs, built into the Microsoft 365 tenant you already pay for. Satya Nadella called Agent 365 Business Analytics "your ERP system for AI usage." Stephen's take: the spend dashboard is a great start if you count hours as capacity, not cash. Let Autopilot run in preview on non-critical work. Expect standalone AI tools to be absorbed into Microsoft, the way video calls and chat were into Teams. Copilot Code apps are a good place to start. And if you are a Microsoft shop, keep what you build elsewhere portable.Available now (GA): Office in Copilot, Copilot Managed Runtime, Microsoft IQ, agent usage metrics and the basic agent registry, Agent 365 and E7. Frontier preview: Business Analytics (50+ Copilot licenses first, all Frontier tenants by about 7 Dec); Autopilot (off by default, US first, billed per use, no GA date); Copilot Code (GA expected around December); Workspaces. Preview: Entra Agent ID for third-party agents, Dynamic Agent Risk. Conflict: Perception is announced GA, but Microsoft's setup page still says limited preview. Autopilot, Code and Cowork use Copilot Credits and need a spending policy.Four steps for this monthAsk IT whether your tenant is in Frontier, and turn it on for a small group. Why: most new features need it, and a small group limits surprises.Set a monthly and per-user spending cap with alerts before anyone uses metered features. Why: the cap comes first.Get one list of every agent, Microsoft and otherwise, each with a named owner. Why: an agent without an owner is one nobody watches.Run one Autopilot pilot on non-critical work and one Copilot Code app that replaces an emailed spreadsheet. Why: learn now, decide on rollout after GA.Watch-outs. Microsoft's value figure is assisted hours times an hourly rate you set. Autopilot conversations are not yet covered by eDiscovery, retention or legal hold. Keep instructions, process steps and data somewhere you own.Quick wrap-up: Perplexity Context API. Send a batch of web addresses, get back clean text plus the passages that answer your question, at about $1 per 1,000 pages, with failed fetches free. It makes research agents cheaper, for example monitoring competitor or regulator pages. Quality figures are Perplexity's own, and your page list reveals what you research. Launch; pricing.Terms. Frontier: Microsoft's early-access program. Agent 365: the control plane for agent identity, security and management. Autopilot: an always-on agent with its own ID, memory and manager.Sources: Nadella keynote; Book of News; Business Analytics; Autopilot requirements; Copilot Code admin; Perception setup.The AI Brief from the YPO Technology Network is a daily executive briefing on the AI developments that matter to business leaders. Hosted by Stephen Forte.
Episode 200. Stephen started this show because almost nobody was talking to CEOs about AI: plenty of tabloid headlines and valuations, little on where the technology is going, what a busy CEO needs to know, and how to gain an edge. He spends two to three hours a day on AI research, and it is perishable, so he shares it. Thank you for helping make this one of the top five percent of podcasts in the world.On 11 November, Databricks CEO Ali Ghodsi showed TechCrunch how he uses his Genie agent: on the weak parts of the employee survey, to rate and discard fixes his executives suggested, and to summarize a sales pitch that went badly. At about 12,000 employees, past Dunbar's number of about 150, the boss hears less directly. Stephen's take: teams rarely filter on purpose. Everyone has a different lens. AI lets a CEO read raw survey and feedback data through their own, program out bias, and focus on the metrics that matter. AI suggests, with sources. People decide.Key takeaways. Use AI on the raw material, not the summary. A wrong definition gives a confident wrong answer. Hours saved are capacity, not cash. Genie and Genie Ontology are generally available. The always-on Genie, Genie Search and Ontology Manager are coming soon to preview and beta. They are not in customers' hands yet.Five steps to start this weekPick one question where you may be seeing only one lens: morale on a team, a slipped deal, what customers ask for. Why: a narrow question gives a result you can check.Give the agent the raw material, not the summary deck: survey comments, lost-deal notes, support tickets. Why: the summary already carries someone else's lens.Write down the definitions it must use (below). Why: it can only answer straight if it knows what your words mean.Tell your team what the agent reads before you start. Check privacy rules and keep survey comments anonymous. Why: finding out later destroys trust.Ask for suggestions with sources, not decisions. Every point traces to a comment or record. Why: you can check it, and a person decides.Starter definitions. Revenue: booked, billed or recognized, and which system is official. Active customer: what activity, over what period. Churn: customers or revenue, monthly or yearly. Official report: which one wins when two disagree. Who can see what: individual comments versus totals.Terms. Genie: Databricks' AI agent for business users. Genie Ontology: the layer holding a company's definitions, metrics, people and processes. Always-on agent: works in the background and comes back for decisions (Genie's is coming soon to preview). Dunbar's number: about 150, the rough limit for stable relationships.Examples named. Ali Ghodsi at Databricks (survey, executive fixes, lost pitch). Stephen uses AI on survey and feedback information to read sentiment through his own lens, with AI suggesting and people deciding.Sources: Databricks release; TechCrunch; Databricks blog.The AI Brief from the YPO Technology Network is a daily executive briefing on the AI developments that matter to business leaders. Hosted by Stephen Forte.
On 13 November 2026, Anthropic published a guide on agents that keep working when nobody is in the chat. Always on is the experience, not staying awake all night. The agent sleeps, wakes for a short run, saves what happened, and goes back to sleep. Hours it saves are capacity, not cash. Unattended agents rarely go rogue. They stall, stop, or break silently. Run an agent attended first, then let it run on its own, and have a second agent check the work. If a team or vendor cannot say how the inbox, timers, memory, permissions, and outbox work, the agent is not ready. The later test: would you know if it stalled overnight.Key takeaways. Always on is the experience, not the runtime. Hours saved are capacity, not cash. Run attended first, then unattended, with a second agent reporting success and failure. Automatic replies are a comfort-level choice. Automatic drafting is a different decision. Start the agent quiet, between over-notification and missing what you need to know.Five building blocks (Anthropic).Event inbox. Every trigger writes to one durable queue, with one active worker. If the worker dies, a new one can retry. Ask: does every trigger go through one queue, and what happens if the worker dies mid-run.Timers. Follow-ups the agent sets for itself, and recurring schedules. Ask: can I see them, and can I turn them off.Memory. Keep only what it cannot look up again. Save a pointer if the fact already lives elsewhere. Ask: what does it keep, where does that live, and can we read it.Permissions. A run holds a person's access. Each action should proceed, block, or wait for a named person's approval. A proxy outside the agent adds the password or the key. Ask: does the agent hold the credentials, which actions wait, and what if that person does not answer.Outbox. The user gets the outcome and a few details, not a live stream of every step. Ask: how does it decide when to message someone, can we start it quiet, and is there a log of sent and unsent messages.Checklist (as spoken). You do not need to write this down as you listen.Run the agent attended first. Then let it run on its own. Why: you learn how it behaves while someone is still watching.Have a second agent check the work and report success and failure. Why: they stall, stop, or break silently, and the job can look like it is still running.Write down which messages can go out on their own, and which wait for a person. Automatic replies are a comfort level. Automatic drafting is a different decision. Why: drafting can go first. Sending is still a separate yes.Ask the five questions above. Why: if a vendor cannot answer them, the agent is not ready to run unattended.Once it is running, ask whether you would know if it stalled overnight. Why: that is a better test than asking whether it ever went rogue.Terms. Always-on agent: feels unattended, but sleeps between short runs. Memory: what is kept between runs, and is inspectable. Credential proxy: inserts the password or key. Outbox: how the outcome reaches a person. Start quiet: fewer messages first, then earn attention over time.Examples named. We run agents unattended after an attended check, with other agents reporting success and failure. They rarely go rogue; the watch is for silent stall. Some messages get automatic replies, many do not. Most emails get an automatic first draft. Drafting is not sending.Sources: Anthropic guide; Elvis Saravia on X.The AI Brief from the YPO Technology Network is a daily executive briefing on the AI developments that matter to business leaders. Hosted by Stephen Forte.



