Your AI Bill Shouldn't Scare You

I Woke Up and My AI Model Was Gone.

Claude launched Fable as its most powerful model yet. Within days, the US government placed it under export controls, and Anthropic had to disable it for every customer to comply. In E15, Jeff uses that overnight shutdown to make the case at the heart of Decision 5 of The High-Stakes AI Playbook. Never hinge your business on a single vendor, a single model, or a single platform. He walks through the three layers of control that keep an operation running when a tool disappears: the tools you choose, the data you own, and the people you prepare. IN THIS EPISODE - Why the Fable shutdown is a real-world example of the Vendor Capture Trap - The strongest case for going all-in on one vendor, and why it does not hold - Layer one: choosing flexible tools you can switch between - Layer two: owning your data so you can move between models - Layer three: preparing your people to handle change without losing a step CHAPTERS 00:00 The morning Fable disappeared 00:32 What the US government actually did 01:06 Decision 5: don't bet your business on one vendor 01:24 A disagreement with a friend in the industry 02:31 The $1,500 cap and what it tells you 03:32 Why I build with model-switching 04:53 Layer 1: Choose flexible, switchable tools 07:45 Layer 2: Own your data 10:36 Amplify, don't replace 12:16 Layer 3: Prepare your people for change 14:26 Take the driver's seat KEY TAKEAWAYS 1. A tool can vanish overnight through no fault of your own, so build for that from the start. 2. Control over your inputs is what protects continuity, from the solo operator to the global CEO. 3. Amplify your people with AI, and never bet the business on one model. LINKS Request the book: goSOS.ai/playbook [https://goSOS.ai/playbook] The Signal on Substack: jeffswan18.substack.com [http://jeffswan18.substack.com] Follow Jeff: @jeffswan18 Stay human boss. Get full access to #NoVendors by Jeff Swan at jeffswan18.substack.com/subscribe [https://jeffswan18.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

You Don't REALly Need AI. Here's When You Should Actually Deploy It.

Episode 14: You Don't Need AI A lot of operators are asking lately whether they really need AI. The honest answer surprises most of them. Nobody really needs AI. It is a tool, and tools serve specific purposes. What most operators get wrong is treating AI as a category instead of as a deployment choice. What you'll hear: - Why AI tools are designed for engagement the same way social media platforms are - The difference between structured automation (deterministic) and AI (probabilistic) - How both technologies combine in the flood mitigation agent we have been building - The Air Canada chatbot case and what unstructured tools do when they manage structured conversations - Why the real failure point in most AI deployments is leadership culture, not the technology - The three questions to answer before deploying any AI The three-question test: 1. What problem are you actually solving? 2. How else could you solve it? 3. Who is going to make it work? Three yes answers mean the deployment is worth committing to. Any no is a signal to slow down. The framework is Decision Three in The High-Stakes AI Playbook, launching Q3 2026. The manuscript is open for early access reads. Reply to the newsletter or DM me on LinkedIn for a copy. Find more at outboundsos.com [http://outboundsos.com] Subscribe to the #NoVendors newsletter at jeffswan18.substack.com [http://jeffswan18.substack.com] Get full access to #NoVendors by Jeff Swan at jeffswan18.substack.com/subscribe [https://jeffswan18.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

Your AI Bill Shouldn't Scare You

Episode 13: Your AI Bill Shouldn't Scare You Most operators looking at AI see the headline pricing and budget for that. What they miss is the way AI tools actually charge, which is by token consumption. Long chatty inputs and detailed outputs both consume tokens, and the bill arrives at the end of the month based on what you used. The fix is not about using less AI. The fix is matching the model you are using to the task you are trying to accomplish. What you'll hear: - Why AI bills surprise operators and what's actually being measured (tokens, not seats) - The three model tiers - premium, workhorse, budget - and how they price differently - A real comparison where the same task cost $1,000 on premium and $42 on budget - Three questions for matching the model to the task (complexity, volume, judgment) - How caching and batching compound the savings - A real example of how SOS Signals uses all three tiers in a coordinated workflow The framework is a compressed version of Decision Four in The High-Stakes AI Playbook. The book covers the full build versus buy decision tree and the architecture choices that determine whether AI is a cost line or a profit lever. The manuscript is open for early access reads. Reply to the newsletter or DM me on LinkedIn for a copy. Find more at outboundsos.com [http://outboundsos.com] Subscribe to the #NoVendors newsletter at jeffswan18.substack.com Get full access to #NoVendors by Jeff Swan at jeffswan18.substack.com/subscribe [https://jeffswan18.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

How To Use AI Without Breaking What Works

The Signal E12: Amplify, Don't Replace Humans Big Tech is selling owners a story about AI that says cut staff, deploy AI, and watch the stock surge. That playbook works for Big Tech. It does not work for businesses where people are on the job site, in the field, in the clinic, or in the truck. This episode walks through three real AI deployments from high-stakes businesses and ends with the three-question test I run before any deployment touches a client company. What you'll hear: - Why the Big Tech AI playbook fails in businesses with hands-on operations - How a live flood intake agent amplifies field assessments without replacing the human walking the property - The mining ops manager's four-question test for any new technology (easier, faster, safer, better) - Why AI-generated cold emails make every seller sound the same, and how to use AI for sales without losing your voice - The three-question test I run before any AI deployment at a client company - Why the People Protection Principle is the rule that holds all of this together The three-question test: 1. Can it deliver without staff reduction in 12 months? 2. Are the people affected the ones carrying institutional knowledge? 3. Have you measured team experience as carefully as financials? Three yes answers means the deployment respects the People Protection Principle. Any no is a signal to slow down. The People Protection Principle is one of seven critical decisions in The High-Stakes AI Playbook, launching Q3 2026. The manuscript is open for early access reads. Reply to the newsletter or DM me on LinkedIn for a copy. Find more at outboundsos.com [http://outboundsos.com] Subscribe to the #NoVendors newsletter at jeffswan18.substack.com [http://jeffswan18.substack.com] Get full access to #NoVendors by Jeff Swan at jeffswan18.substack.com/subscribe [https://jeffswan18.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

I Wrote a Book About AI and People. It's Time to Send the Manuscript.

Big Tech is telling owners they have to build with AI or get left behind. That story is wrong about how most real businesses actually work. This episode opens with three patterns from a year of building AI agents alongside hands-on owners and CEOs of high-stakes companies. - AI moves fast (and breaks things every model update). - Specificity hurts AI performance (which is why knowing when to deploy generative versus deterministic software matters). - Hallucinations make blind trust dangerous (which is why every Signal we build runs through human verification). Then the announcement. I have written The High-Stakes AI Playbook, covering seven critical decisions hands-on operators need to make before deploying AI. Targeting Q3 2026 launch. The book is for owners and CEOs of high-stakes businesses with real choice about how AI fits into their operations. It is not for companies in declining industries making survival decisions, or public-company CEOs operating under short-term fiduciary mandates. The manuscript is going out to a thousand operators, analysts, and commentators. Honest reviews from real operators may end up in the foreword. What you will take away: - Three patterns that surface in every AI build for high-stakes operators - A framework for when to deploy generative AI versus deterministic software - The case for keeping humans accountable in the loop - How to get a copy of the manuscript Your move: If you are running a high-stakes company and want a copy of the manuscript, watch your inbox, reach out by DM, or email. Honest feedback is what I am asking for. Find more at outboundsos.com [http://outboundsos.com] Newsletter at jeffswan18.substack.com [http://jeffswan18.substack.com] Get full access to #NoVendors by Jeff Swan at jeffswan18.substack.com/subscribe [https://jeffswan18.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]