Field notes on websites that answer, bookings, and local search

August 13, 2026·16 min read

Agents do not fail in the model. They fail at the data perimeter — and that is why governance vendors matter again.

Production agents need an enforced data perimeter: FGAC, lineage, observability, and drift detection in the tool path — not prompt guardrails alone. Why data-access platforms inherit the architecture LLM-security startups must bolt on.

August 12, 2026·8 min read

How to run AI in your marketing campaigns without becoming a content mill

A campaign operating layer for mid-market teams: brief → draft → human edit → distribute → measure. What to automate, what to keep human-only, and a 90-day starter plan that doesn't require faking your numbers.

August 10, 2026·9 min read

How to burn fewer tokens in Cursor and Claude without making the work worse

Practical token discipline for operators using Cursor and Claude daily: what actually burns spend, scoped context, prompt caching, agent-loop hygiene, and team rules that cut the bill without cutting quality.

Data

Gartner: 60% of AI projects will be abandoned over data that was never AI-ready. Here's what 'AI-ready' actually means.

Gartner predicts organizations will abandon 60% of AI projects unsupported by AI-ready data. Cloudera's 2026 numbers show the Readiness Illusion: companies claim strategy while access and governance lag. Operator translation inside.

July 24, 2026·4 min read

Gartner's first AI-governance Magic Quadrant is excellent. It's also priced for someone else.

Gartner's inaugural AI Governance Platforms MQ (June 2026) prices the category for enterprises — $65M today, $1.4B by 2030. Below the Governance Poverty Line, obligations still apply. Here is what mid-market operators can actually assemble.

July 20, 2026·6 min read

Gartner says 40% of agentic AI projects will die by 2027. The autopsy is more useful than the number.

Gartner's 40% agent cancellation forecast and MIT NANDA's 95% zero-ROI claim dominate the failure discourse — but the evidence is weaker than the headlines. We call it the Scoping Crisis: management failure, not model failure.

July 15, 2026·9 min read

What AI agent implementation really costs in 2026

AI agent implementation cost in 2026: phase-by-phase breakdown, hidden costs, and illustrative inference ranges. Background note — current work is web packages and US→EMEA.

AI agents vs RPA vs Zapier: an honest comparison for operators

AI agents vs RPA vs Zapier in 2026: capability, cost and maintenance compared honestly — when the boring tool wins, and when an agent is genuinely the fit.

AI for independent hotels: a practical guide (Spain & Southern Europe)

AI for independent hotels in Spain and Southern Europe: multilingual guest triage, draft replies with human approval, seasonal scale — no invented connectors. Current work starts with a website package.

AI process automation: a complete guide for operations teams

AI process automation in 2026: what to automate first, illustrative weekly hours table, four ROI flows, and human oversight on sensitive actions. Background — current work is web packages and US→EMEA.

July 12, 2026·6 min read

How to automate customer replies with AI — without losing the human

Automate customer replies with AI in seven steps: triage, templates, MCP tools, open/frontier routing, and a human approval queue before every send.

July 12, 2026·2 min read

What getting Claude-certified taught us about building real AI systems

We build on open models by default — and just got certified on the most premium stack in the industry. Here's why that isn't a contradiction, and five lessons from Anthropic's Claude Certified Architect exam that apply to any operator putting agents into production.

July 12, 2026·4 min read

The EU AI Act for operating companies: what actually applies to you now

EU AI Act for operators after the Digital Omnibus: Article 50 transparency 2 Aug 2026, Annex III high-risk 2 Dec 2027 — what normal businesses using agents must do and ignore.

Open-source vs frontier models: the real cost math for business automation

Open vs frontier LLM cost for automation in 2026: routing logic, illustrative per-call economics, ~80/20 blended savings, residency, and when premium APIs earn their place.

What is an AI Operating Layer? (And why your next hire might be one)

An AI Operating Layer connects governed agents, your data, and hybrid models under one stack — open by default, human oversight built in. Background note — current work is web packages and US→EMEA.

Your model stack just became a governance decision

Grok 4.5's EU gap, GLM-5.2's open weights, and a flagship model priced at double its predecessor: three launches in one month that turned model choice from an engineering preference into a cost and data-residency decision.

July 9, 2026·5 min read

The AI Act 'delay': a reprieve or a trap?

The Digital Omnibus moved the headline deadline — but transparency obligations, Spain's draft AI law, and your supervisor didn't move. A practitioner's reading.

July 8, 2026·2 min read

The Omnibus moved your deadline. Here's what to do with the runway.

The EU Digital Omnibus deferred high-risk AI Act duties — but Article 50 transparency still lands in August 2026. A practical plan for the months you actually have.

July 5, 2026·6 min read

Data governance without the headache

How small and mid-size companies can get privacy and records right early — turning compliance from a tax into an advantage.

July 1, 2026·2 min read