Lessons from building GTM engines and AI agent systems in production — what works, what breaks, and why.
Twelve months building the automation layer behind Linguana's outbound, onboarding and support. Two thirds of outbound now sent without a human writing it, ~600 tickets a month closing on their own — and the chart that proved we'd been scaling the wrong decision.
Read post →Gartner expects 40% of agentic AI projects to be canceled by 2027 — and not because the models are bad. The best lead-capture system I ever shipped was a spreadsheet, a sync, and a tag.
Read post →I built a sales agent to answer leads faster. Then I sat behind the reps and watched them work — and found out I'd automated the two-minute part of a forty-two-minute job.
Read post →It's a QA metric wearing a business metric's clothes. It goes up when you make the agent worse, and it cost me weeks. What I report instead.
Read post →Your AI agent isn't misbehaving — your prompt is negotiating. Why "prefer X when…" fails in production, and the decision-tree structure that fixed our misclassifications.
Read post →Leads that already agreed in principle were going cold in an inbox. Here's the autonomous agent we built to close the gap — and how it now closes 35% of contracts without a human touching them.
Read post →Every GTM stack I audit has the same disease: ten tools, zero system. The difference between a stack that costs money and an engine that compounds.
Read post →30 minutes. Bring your messiest workflow — I'll tell you what I'd build.
Book a 30-min Meeting