— Writing

Essays, takes,
and field notes.

Written for operators running real businesses, not for the content-marketing treadmill. What I've actually used, what broke, what I'd do again.

— JUL 17, 2026

Open weights closed the gap. Nobody closed the safety gap.

Thinking Machines shipped free frontier weights. GPT-5.6 deletes files. Cursor's RCE sat unpatched for months. This week's signal, compressed.

signaldigestopen-weightsai-safety

— JUL 14, 2026

Grok Build CLI leaked repos. Here's the identity gap.

xAI said nothing left your machine. A wire-capture proved otherwise. Here's how real agent identity — scoping, sandboxing, audit trails — actually works.

craftai-agentssecuritymcp

— JUL 10, 2026

Agents got cheaper this week. The guardrails didn't keep up.

Grok 4.5 undercut Opus pricing by 60% this week — and three unrelated stories show benchmarks, permissions, and security all lagging behind.

signaldigestai-agentssecurity

— JUL 7, 2026

Your AI coding tool phones home. Alibaba just proved it.

Alibaba banned Claude Code over hidden telemetry logging timezones and proxy data. Here's the vendor-trust audit most SMB operators never run.

authorityai-vendor-trustclaude-codevendor-audit

— JUL 3, 2026

Four AI labs just committed $9B to babysit your rollout

This week: Microsoft topped Amazon's AI deployment bet by 2.5x, a BU study shows 'AI employee' framing hides errors, and Sonnet 5 undercuts Opus on price.

signalai-agentsenterprise-aianthropic

— JUN 30, 2026

Operators using AI hardest are pulling ahead. Data confirms.

Workers delegating the most to AI are the most optimistic about their future. Anthropic's data is out. Here's what it means for SMB operators.

authorityai-automationoperationssmb

— JUN 26, 2026

Your AI vendor wants your passport. This week's signal.

Anthropic added government ID checks. Robots beat human engineers 37×. DeepMind lost its AlphaFold architect. What the week's signal means for operators.

signaldigestai-toolssmb-ops

— JUN 19, 2026

ChatGPT slipped below 50%. That's not even the biggest news.

ChatGPT dropped below 50% MAU for the first time. Anthropic took the enterprise spend lead. SpaceX bought Cursor. This week, every metric moved at once.

signaldigestai-marketchatgpt

— JUN 16, 2026

Your AI stack can be recalled. Fable 5 just proved it.

On June 12, the US government suspended access to Anthropic's flagship models overnight. No warning, no patch window. Here's what operators need to rethink.

signalai-infrastructurevendor-riskanthropic

— JUN 12, 2026

Fable 5 is live. The AI productivity math doesn't add up.

Glean says AI saves 13 hours but costs 6.4 in overhead. Fable 5 raised the ceiling anyway. Both major API vendors are now on the IPO clock. June 12 digest.

signaldigestfable-5anthropic

— JUN 9, 2026

Most operators still generate reports by hand. I don't.

How I automated a client's 8-hour weekly ops report using Claude Code. Token cost: $20/month. Ops time replaced: 400 hours/year. Here's the exact workflow.

craftclaude-codeautomationplaybook

— JUN 5, 2026

Anthropic filed its S-1. Here's what else moved this week.

Seven signals from the week ending June 5: an IPO filing, an AI spend wall, and infrastructure bets that shift the calculus heading into Q3.

signaldigestanthropicai-tools

— JUN 2, 2026

Your ops are fine—until the next growth sprint breaks them

Five operational warning signs that surface right after a revenue spike — and what each one tells you about where your next ceiling is hiding.

authorityoperationssmbautomation

— MAY 29, 2026

Claude Opus 4.8 ships today. Your flat-rate AI deal is next.

Claude Opus 4.8 lands with 4x better code-flaw detection. Anthropic and OpenAI ended flat-rate seats. Remote.com published the AI ROI receipts.

signalclaudepricingsmb

— MAY 26, 2026

Token prices are falling. Your AI bill isn't. Here's why.

HBM memory costs rose from 52% to 63% of AI chip spend in one year. The Jevons paradox and a gas-first infrastructure bet mean your cost model needs a rebuild.

signaladjacentai-costsinfrastructure

— MAY 22, 2026

npm hit Claude Code's hooks. I/O changed everything else.

npm attackers targeted Claude Code's SessionStart hooks directly. Google I/O crossed 1B AI Mode users. Anthropic profitable — compute costs explain why.

signalweekly-digestgoogle-iosecurity

— MAY 19, 2026

AI agents pass demos. They fail 96% of real SaaS work.

SaaS-Bench tested frontier agents across 106 real workflows — completion under 4%. Here's what that means for operators building on AI agent infrastructure.

signalai-agentsbenchmarkssmb-ops

— MAY 15, 2026

Anthropic passed OpenAI. Who else shifted this week.

Anthropic flipped past OpenAI in business payments. n8n landed SAP. AI vendors started competing with their own API customers. Seven shifts, operator framing.

signaldigestanthropicn8n

— MAY 12, 2026

Claude knows it's being tested. Your evals are optimistic.

Anthropic's NLA research found Claude silently detects eval conditions 16–26% of the time. Here's what that means for your automated quality gate.

craftai-agentsevaluationanthropic

— MAY 8, 2026

Default changed, ceiling raised: this week's signal

This week: GPT-5.5 is the new ChatGPT default, Anthropic doubled rate limits, a 45x agent cost gap got benchmarked, and DeepSeek went state.

signaldigestgpt-5anthropic

— MAY 5, 2026

AI agents can buy from your store now. Your checkout can't.

Stripe Sessions 2026 shipped the payment layer for AI agents. Here's the 20-minute audit every SMB operator selling digitally should run this week.

authoritystripeai-agentssmb-ops

— MAY 1, 2026

Agent plumbing just got serious. This week's signal.

Symphony, AutoAdapt, and multi-agent red-team findings all landed this week. Plus a Microsoft 365 Copilot pricing deadline SMBs shouldn't sleep on.

signalai-agentsorchestrationmicrosoft

— APR 24, 2026

Cloud Next, GPT-5.5, and the model Anthropic won't release

This week: Google's Cloud Next bet $750M on agent infrastructure, OpenAI shipped GPT-5.5, and Anthropic confirmed their best model won't be public.

signaldigestai-agentsopenai

— APR 22, 2026

Opus 4.7 Made Agentic Work Accountable. Here's How.

Opus 4.7 shipped April 16. xhigh effort default in Claude Code, task budgets public beta, vision 3×. The 48-hour operator read on what matters.

craftclaude-codeclaudeai-agents