In partnership with
I had been giving instructions to five subagents for months. None of the five existed. How does a setup keep working with an imaginary team?
This week
I've made peace with Claude since Fable 5.1. In my day-to-day use, it no longer burns through my credits in twenty minutes. I can get back to work without constantly watching what's left. That's my experience, not a comparative measurement of quotas.
But that little affair introduced me to Codex. And it really holds its own. It handles code, of course, but also media and applications. That range matters when a project needs more than a TypeScript file. Claude has earned its place back. Codex is keeping its own.
With two agents, I also have two setups to maintain. When I tried claude plugin eval init, I ran it from an application repo instead of a plugin folder. The error made me look at what was actually installed. My audit found five missing subagents, seven plugins installed twice, and a permission rule that didn't protect the files it targeted. All of that coexisted with sessions that appeared to work normally.
Before fall, I'd set aside thirty minutes for this cleanup. In Claude Code, start with /doctor, then check your active plugins and MCP server status. Make sure an agent named in your instructions actually exists, and that a hook still points to a script that's there. In Codex, inventory the plugins, skills, and connections you use in one real project. A tool installed for an experiment six months ago should earn its place.
Keep shared instructions in one source, and reserve specialized instructions for the projects that need them. After each removal, open a fresh session and rerun a small, familiar task. My choice for this fall is to keep both agents, remove the duplicates, and check what's actually doing the work.
Hire Ava, the AI BDR built for enterprise
Ava is the first AI BDR to run outbound end to end, and you decide whether she runs autonomously or on copilot.
She finds leads or ingests accounts from your CRM, enriches them, sends personalized emails on behalf of your reps, follows up, handles replies and books meetings. Website visitor de-anonymization, intent signals, and a parallel dialer come built in.
A small team can manage her centrally for thousands of reps who never log in. Everything syncs two-way with Salesforce and HubSpot.
Ava runs outbound for companies like DoorDash and Grammarly, and one customer deploys her across 1,000+ reps. Ava is SOC 2 Type II audited, SSO and GDPR ready. Ava is how revenue teams grow pipeline without growing headcount.
In the news
Claude Code plugin evals let you compare the same case with and without a plugin. claude plugin eval init prepares the suite from the plugin root; it doesn't clean up your entire installation. If results are just as good without the plugin, that test hasn't demonstrated its value. That's a question worth asking of extensions we keep out of habit.
This week on the blog
Anthropic Shipped Plugin Evals. The Error It Threw Found 5 Agents That Never Existed. The full audit covers the ghost agents and how I organized skills across Claude and Codex.
Revenue Stack: September 2026
Before handing more work to an agent, you need to give it a precise result to aim for. That's what my book Vibe Coding, For Real covers: framing the work before you start building.
Phil
My imaginary team held together because delegations fell back to a generic agent. I was getting answers, but not from the team I thought I'd configured. That's the gap I want to close before adding the next plugin.
PS: which plugin did you install and then completely forget about?
You're receiving this because you signed up at rentierdigital.beehiiv.com.