Support triage at a Series B SaaS
How we cut median first-response time 64% on a Zendesk workflow without removing the human-approval gate.
−64%
FRT
−38%
Cost
6 wk
Shipped
ResourcesBuild logs, guides,
Build logs, guides,
and downloads.
Everything we publish about shipping production AI agents. No gated walls except the checklist (which we'll happily trade for your email).
Featured downloadThe 42-Point AI Agent
The 42-Point AI Agent
Workflow Readiness Checklist
Score your stack across routing, governance, evals, cost control, and human-in-loop. PDF + Notion template. Used in every audit.
No spam. Unsubscribe with one click.
readiness_checklist.pdf42 items · v3.1
Every agent has a named owner
Tool access scoped per agent
PII redaction policy defined
Per-route monthly budget set
Fallback model chain documented
Eval suite runs on every PR
Human approval for irreversible actions
Full event log retained 90+ days
...34 more
Build logsRedacted ship logs
Redacted ship logs
from recent engagements.
Every client engagement ends in a published log: what we built, what we tried that didn't work, the numbers, and what we'd do differently.
Research brief generator for a 40-person agency
Multi-step research agent with citation gates that replaced 2 FTE of manual prep. Every claim sourced; editors love it.
9.2×
Throughput
98%
Cite rate
4 wk
Shipped
Governance retrofit on 30 n8n workflows
A Series C fintech had production agent flows, zero audit log, and no PII policy. We added the gateway without rebuilding a single flow.
100%
Audited
−52%
Token spend
3 wk
Shipped
Lead routing for a B2B marketplace
Replaced 3 different Zapier flows with one routed agent. SDR triage time dropped 71%; CRM owner accuracy hit 96%.
−71%
Triage
96%
Accuracy
5 wk
Shipped
Knowledge assistant over 1,400 Notion pages
Permissioned RAG that respects Notion's sharing rules. Per-user retrieval scope; weekly drift evaluation.
94%
Accept
0
Leaks
4 wk
Shipped
Sales follow-up for a 12-rep team
Context-aware drafts in the rep's voice, with compliance check before every send. Reply rate up 3.4×, opt-outs down 22%.
3.4×
Reply
−22%
Opt-outs
6 wk
Shipped
GuidesLong-form pieces
Long-form pieces
on routing & governance.
The agent-routing stack, 2026 edition
A walkthrough of Portkey, OpenRouter, LiteLLM, and Helicone — when each makes sense, and when to roll your own.
Read →MCP vs. A2A: a working taxonomy
How we think about agent-to-agent vs. agent-to-tool protocols, with code examples and migration patterns.
Read →Evals that actually catch regressions
The eval patterns we use on every Build Sprint — synthetic, replay, and live-traffic, with concrete examples per workflow type.
Read →Cost guardrails for multi-model setups
Per-route budgets, model selection heuristics, and the cache hit-rate math that saved one client $14k/month.
Read →From Zapier to gateway in 7 days
The migration playbook for teams whose AI lives inside Zapier or Make today — and what's worth keeping there.
Read →Human-in-loop design patterns
Where to put approval gates so they actually catch errors, without turning your team into a Mechanical Turk for the LLM.
Read →Free templates
Steal these and ship.
AI Workflow Inventory Template
The intake form we use on every audit. Run it on your own stack to surface what you actually have.
Get the template →routes.yaml starter
A working RouterA2A config with sensible defaults: fallbacks, budgets, redaction, and rate limits.
View on GitHub →Multi-model cost calculator
Plug in your traffic and model mix, see projected spend, savings from routing, and break-even on a gateway.
Copy the sheet →Read the resources.
Then book the audit.
The shortcut to all of this in your stack, with a written roadmap to ship.