Episode Summary

Dave now runs two AI personal assistants, one for his life and one for App Vitals, and on the drive in he realized they had started talking to each other in Slack. Dave and Dan dig into what that says about where agents are headed: one assistant you talk to, backed by specialist agents that each own a narrow job and their own credentials.

▶ Subscribe to The Velocity Lab on YouTube: youtube.com/@app-vitals

Key Topics

• How Shipwright thinks about an "agent": a level above sub-agents, with memory, crons, skills and plugins
• Two failed attempts at autonomous, KPI-driven PRD agents, and why the third try pulls the core agent out of Shipwright
• Why Dave talks to his assistants by voice, and why a unique voice makes him trust it more
• An assistant that reads the cloud bill and quota warnings every day, with read-only access and no secrets
• Where to draw the line on an agent's scope: function and access as the two dimensions of a specialist
• Why focused specialists keep cleaner memory than one assistant fielding a million requests a day

Notable Quotes

• "Both my personal assistant and my App Vitals assistant are both in Slack, and now they are starting to communicate with each other."
• "If I talk to something that has a unique voice, I trust it more."
• "Function and access are the two differentiators for a specialized agent."

About The Velocity Lab

Dave O'Dell and Dan McAulay work inside engineering organizations every day helping them ship faster with AI. No hype, no BS — just what's working in the field.

Subscribe: RSS

Episode Summary

Teams evaluating Shipwright keep asking the same question: why only Claude Code? Dave and Dan walk through their evaluation of the Agent Client Protocol (ACP), what it supports today, where it falls short for autonomous work, and why they think token cost is the last problem an engineering org should be solving.

▶ Subscribe to The Velocity Lab on YouTube: youtube.com/@app-vitals

Key Topics

• What ACP is and where it sits in the stack, between your chat or editor and the coding agent CLI
• What ACP handles well today: sessions, prompting, streaming and tool calls
• The gaps that matter for autonomous coding: skills, sub-agents, hooks, budgets and reliable cost data
• Why switching CLIs means rewriting skills, and why skills even need tuning when the model changes
• The two real reasons to go multi-model: cost on self-hosted open models, and security for data that can't leave your network
• Why getting 2X throughput from autonomous agents comes before optimizing the model bill

Notable Quotes

• "Token cost, trust me, is the last problem you need to solve."
• "Your code review step is not gonna be 2X better because Claude wrote the PR and Codex reviewed it."
• "Fix the throughput. Don't change the model."

About The Velocity Lab

Dave O'Dell and Dan McAulay work inside engineering organizations every day helping them ship faster with AI. No hype, no BS — just what's working in the field.

Subscribe: RSS

Episode Summary

Two years ago an AI mistake meant a misformatted screen. Now it can mean a security leak or a nuked database. Dave and Dan explain why agent security is the new DevSecOps, and why an autonomous agent in an isolated cloud environment can actually be less risky than a human running one on their laptop.

▶ Subscribe to The Velocity Lab on YouTube: youtube.com/@app-vitals

Key Topics

• Why the stakes changed: agents now run autonomously with access to real systems
• Model-level safeguards are improving, but models are still probabilistic, so infrastructure has to carry the load
• Blast radius: an agent can only touch what you give it, and a token without merge rights cannot merge
• Why a sandboxed cloud agent can be safer than local agents on laptops with elevated permissions
• Lessons from working with PCI clients: start with zero permissions and grant only what is needed
• Planning in the loop locally vs. autonomous execution in the cloud

Notable Quotes

• "It's actually less risky to have an autonomous agent running in the cloud writing software than having it running locally."
• "If you don't give it the ability to merge a pull request, it can't merge a pull request. Black and white."
• "Everyone's gonna be using agents. Just be smart about what you give it access to."

About The Velocity Lab

Dave O'Dell and Dan McAulay work inside engineering organizations every day helping them ship faster with AI. No hype, no BS — just what's working in the field.

Subscribe: RSS