Venture Crane

Articles

96% Token Reduction - Lazy-Loading Agent Context

Feb 7, 2026 · 8 min read

How we cut session startup token consumption by 96% by switching from eager document loading to an index-and-fetch pattern.

From Monolith to Microworker - Decommissioning the Relay

Feb 5, 2026 · 9 min read

We deleted a 3,234-line Cloudflare Worker, its database, and its storage bucket. Here is what we learned about scope creep in serverless architectures.

Staging Environments for AI Agents

Feb 3, 2026 · 10 min read

A 4-phase environment strategy for AI agent infrastructure - Cloudflare splits, automated CI/CD, scoped secrets, and agent-aware routing.

Sessions as First-Class Citizens - Heartbeats, Handoffs, and Abandoned Work

Jan 31, 2026 · 14 min read

Why we gave AI agent sessions heartbeats, idempotent handoff storage, and a full lifecycle - the same reliability patterns used in distributed systems.

Building an MCP Server for Workflow Orchestration

Jan 29, 2026 · 14 min read

A practical walkthrough of building an MCP server that bridges an AI coding CLI to a custom backend API, with lessons on tool design and fleet deployment.

Secrets Injection at Agent Launch Time

Jan 27, 2026 · 12 min read

How a CLI launcher scans repos, matches them to projects, and injects the right secrets without .env files or hardcoded credentials.

Fleet Management for One Person

Jan 24, 2026 · 9 min read

How a solo founder manages a distributed dev fleet with Tailscale, idempotent bootstrap scripts, SSH mesh networking, and macOS hardening.

One Monorepo, Multiple Ventures - Registry-Driven Multi-Tenant Infrastructure

Jan 22, 2026 · 10 min read

How a JSON venture registry with capability flags lets a single monorepo serve multiple products without infrastructure sprawl or cross-contamination.

Multi-Agent Team Protocols Without Chaos

Jan 20, 2026 · 12 min read

How we coordinate dev agents, PM agents, an advisor, and a human captain using namespaced labels, QA grading, and explicit role boundaries.

Kill Discipline for AI Agent Teams

Jan 17, 2026 · 10 min read

How mandatory stop points prevent the most expensive failure mode in agent-assisted development - silent churn on unsolvable problems.