THE THINKING
The methodology, applied.
The same commercial and GTM problems keep showing up. These are the patterns, named plainly, with the fix attached.
Featured this week
Two pieces worth a Wednesday.
The Authenticity Moat: Why Manual Outreach Still Closes (and When To Automate)
Manual outreach is a founder personally sending individual, non-automated messages to potential customers. It still closes better than automated sequences at the early stage, and not because of effort: trust and relationship-building work differently when you're still discovering what converts.
Claude for Small Business closed the integration gap - but only incrementally
Mid-May, Anthropic embedded Claude directly into QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365, packaging workflows for payroll, invoicing, sales, design, marketing, and month-end close. The internet spent the weekend losing its mind about the new Claude for Small Business.
AI insights
Claude Code for Non-Technical Founders: How to Build Without Writing Code
Claude Code builds software from plain-English instructions, so non-technical founders can ship deployed web apps without learning to code.
The AI Startup Failure Report: Why Most AI Startups Won't Survive the Next Model Release
About 40% of AI startups launched in 2024 shut down within two years, but the dividing line isn't model quality, it's whether the company controls something foundation labs structurally can't ship: proprietary data, an embedded workflow, a regulatory path, or a distribution surface. The test: if a stronger model shipped tomorrow, would your product get stronger or weaker? AI didn't change why startups fail; it compressed the timeline.
AI advertising disclosure: what ChatGPT's ad platform means for consumers and founders
ChatGPT's ad format integrates sponsored results into AI responses. The Trust Transfer Problem explains why the channel works short-term and why founders should build genuine audience relationships before the regulatory window closes.
Agentic AI Is Killing Per-Seat SaaS Pricing
Per-seat pricing was always a proxy for value, and AI agents break it. Seat-based SaaS revenue is projected to fall from 21% to 15% of the market by 2030, so price the outcome your software produces, not the seats it's accessed from.
Claude for Small Business closed the integration gap - but only incrementally
Claude for Small Business removes the integration tax that blocked AI adoption in small businesses for two years
How To Run a Four-Agent Dev Team as a Non-Technical Founder
A four-agent team (Designer, Engineer, Reviewer, Marketer) that hands work between each other inside one Claude Code conversation. A non-technical founder can ship a production website, including the headless CMS install, by filling in four context files and giving briefs. Tinctu.re was built this way.
Bixonimania and AI accuracy: why format beats truth
LLMs don’t check truth; they trust patterns of authority. Bixonimania, a fake disease treated as real, shows how format can override accuracy.
How to build an AI Reddit monitor for under $5 a month
A four-dimension scoring engine surfacing 30 ICP-fit founder threads a day, drafting peer-voice comments on demand, running on Anthropic Haiku for $4.50/month.
Claude Opus 4.7: What It Means for Founders
Claude Opus 4.7 checks its own work and follows instructions precisely, so you can trust it with real multi-step business workflows
AI in Revenue Operations for SaaS: Where to Apply AI in Your Commercial Layer Without Wasting Time
Apply AI to high-repetition, low-judgment commercial tasks only after the underlying process is understood.