7 min read
An agent box: where your agent loops actually live
Loop engineering needs somewhere to live: an agent box on your server —an outbound daemon, your own worktree engine and access without opening ports.
#fde#agents#harness
Blog
Essays.
Notes from real engineering work — Forward Deployed Engineering, AI-native vertical SaaS, domain immersion (accounting, finance), and the trade-offs we don't usually write down.
-
- 4 min read
Your role isn't your title: five archetypes for agent-augmented teams
When the agent absorbs execution, what sets you apart stops being your title —engineer, design, PM— and becomes your posture in the product lifecycle.
#ai-digest#fde#teams - 5 min read
Loop engineering: the loop you're not in
Loop engineering isn't 'design loops, not prompts.' It's three decisions: when to raise or lower a loop, how to let it go, and what tool it's missing to close itself.
#ai-digest#agents#harness - 8 min read
The verification ladder: when the domain writes its own code
When the domain generates its code with AI, reviewing it doesn't scale. The verification ladder: let the cheapest verifier check each claim.
#fde#fintech#guardrails - 5 min read
Build your domain benchmark before you build the agent
Ramp evaluates its accounting agent against 237 tasks and 3,469 accountant-written criteria. Why the private domain benchmark decides more than the model.
#ai-digest#evals#fintech - 5 min read
Four boundaries for an agent with access to your books
An agent that reconciles invoices and posts to your ledger holds the keys to your bank. The four boundaries that decide whether a prompt injection can move money.
#ai-digest#guardrails#agents - 6 min read
Your accounting agent isn't incapable, it's unreliable
Aggregate pass@1 falls from 76% to 52% as tasks get longer. In an agent that posts to your ledger, that gap is mis-stated entries. Capability isn't reliability.
#ai-digest#evals#agents - 6 min read
Three parts, one agent: Phil Schmid's 2026 stack
In 22 days, Phil Schmid published three deep dives — skills, MCP, subagents — that read together are the unwritten manual for building a serious agent in 2026.
#ai-digest#agents#harness - 4 min read
Three proofs the harness matters as much as the model
Firefox found 13x more bugs, Zenith won 5 of 8 tasks at 43% cost, Eugene Yan shared his workflow. The signal: invest in your harness, not the model.
#ai-digest#agents#harness