I run paid media for a living and I spend my nights wiring AI agents into marketing workflows. This site is where I test the tools, publish what actually works, and skip the hype.
Latest posts
- What readers do when they smell AI writing: the numbers behind the revolt
Cantrill's reader-revolt post hit the HN front page with survey numbers behind it: 78% of developers stop reading when they suspect LLM text. What the evidence says and the workflow that survives it.
- Agent token costs: what the 90% cut actually takes
Spotify's shunt plugin routes agent grunt work to a cheap model and cut Claude Code token usage 90%. What transfers to your stack, and what stays expensive.
- Human-in-the-loop gates that actually work
Agent approval gates that hold up in production: tool-layer rules, blast-radius tiers and named sign-off, from real builds. See the patterns that work.
- Local-first AI: for teams that cannot send data to APIs
Local-first AI in 2026: one always-on machine, an OpenAI-compatible server, MoE models and a private mesh. The stack, the RAM math, and the honest limits.
- Anthropic vs OpenAI marketing workloads: the September read
Both vendors repriced in five weeks: Fable 5.1 held $10, Sol fell to $4, and GPT-6 Astra reset the top. See which stack fits marketing work.
- llms.txt: what actually moves AI citations
Ahrefs logs show 97% of llms.txt files are never fetched and Google says AI search does not need the file. See what actually moves citations.
- Context engineering kits: inside the 1,652-star repo
What the 1,652-star context engineering kit actually does, what it costs in tokens, and the proof problem. Read the checklist before you install.
- AI agent memory: what agentmemory gets right and wrong
agentmemory is the 28k-star memory layer for AI agents. What it captures, what its benchmarks actually show, and where it falls over, as of September 2026.
- AI search citations: inside the farm Perplexity cites
Three sites published 215,128 machine-made buying guides, and Perplexity cites them. See what the audits measured and how to weigh AI citations.
- Claude Fable 5.1 release notes: what marketers should check
Claude Fable 5.1 keeps $10 and $50 prices and cuts cache reads to a quarter. Three API changes can break a migration. Check the notes before you ship.
- AI in search: what changed for marketing teams
Google merged AI Overviews and AI Mode into one flow and moved Search campaigns to AI Max on September 1. See what changed, as of September 2026.
- Meta AI ad updates: what marketing teams should check
Meta shipped three AI ad changes this summer: an ads MCP server, Muse Image in Advantage+ creative, and text that edits itself. Check your account.
- AI briefs: the workflow that holds up
AI briefs fail when they carry no decisions. I share the brief structure, five inputs that make a draft usable, and the review gate, as of September 2026.
- Meta AI ads 2026: what automated bidding does and hides
Meta's automated bidding optimizes for delivery, not your CPA. Here is what Highest Volume and Advantage+ campaign budget do, and what the interface hides.
- Building marketing agents with Make.com: what we learned
We built marketing agents in Make.com and learned where the canvas helps, where the credit meter bites, and what a real stack costs in 2026. Dated and sourced.
- Performance Max in 2026: what changed for media buyers
Performance Max got real controls in 2026, then the bidding math changed. What to check before August 17 and September's language cut, dated and sourced.
- EU AI Act Article 50: the deadline marketers missed
EU AI Act Article 50 applies since 2 August 2026 and covers more than high-risk AI. Four transparency obligations hit marketing teams now. Read the checklist.
- AI Overviews: the content workflow that holds up
AI Overviews appear on half of tracked queries, and citations no longer follow the top 10. Run the loop: track, structure, answer, refresh, as of September 2026.
- ChatGPT Ads: what media buyers need to know
ChatGPT Ads went self-serve in the US in May 2026 and launched in the UK in August. What it costs, how targeting works, and what is still missing.
- Claude Opus 5: what it changes for marketing teams
Anthropic shipped Claude Opus 5 at Opus 4.8 prices with near-Fable-5 performance. Check the tier math before your next build, as of September 2026.
- GPT-5.6 for marketing teams: what the new model changes
OpenAI shipped the GPT-5.6 family in July, then cut Luna prices 80% in August. See which tier fits your workload, as of September 2026.
- Building marketing agents with MCP servers: what we learned
We built marketing agents on MCP servers. Here is what held up, what failed, and the approval gates that saved us, dated September 2026.
- TikTok AI creative tools: what's free, what changed in 2026
TikTok's AI creative tools changed twice in 2026: Symphony Agent in July, Seedance 2.5 in August. What each costs and where the free tools fall short.
- Google AI ads in 2026: what changed for media buyers
Search ads go AI-native in 2026: AI Max auto-upgrades hit September 1, ads run inside AI Mode, and AI creative has new disclosure rules. Check your account.
- Meta AI creative in 2026: what changed for media buyers
Meta's AI creative stack changed three times in 2026: delivery reads creative pre-auction, generation moved into Ads Manager, disclosure grew teeth. Audit now.
- Automated bidding in 2026: what changed for media buyers
Google changes target-based bidding on August 17, 2026. Budget-limited campaigns will finally hit their targets, so accounts that overperform need action now.
- Programmatic SEO: the workflow that survives Google
Programmatic SEO still works in 2026, but template-swap version is dead. Here is the data-first workflow I run, the gate that protects it, and where AI fits.
- Advantage+ in 2026: what changed for media buyers
Meta made Advantage+ the default in 2026. Targeting and attribution changed with it. What moved, and what to check before your next launch.
- Building marketing agents with n8n: what we learned
Building marketing agents with n8n: what held up, what failed, and what a stack costs in 2026. Dated and sourced, before you build.
- ChatGPT for marketing: the honest test
What ChatGPT is actually good for in marketing in 2026: verified pricing, documented limits, and workflows that survive real campaigns. Read the honest test.