Manager Visibility via an Always-On Agent
Enlist a standing AI session across all repos and channels to stay on top of 8x throughput and drive conversations.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 4
- Confidence
- 88%
To manage when the team ships far more than any human can track, Fiona Fung runs a standing Claude Code session enlisted in all repos with access to Slack and metrics. Instead of only generating PRs and bug fixes, she uses it to review what shipped, surface themes across incidents, and turn those into conversations with the people she supports. Automated 'routines' then run these rituals (e.g. summarizing feedback channels each morning) and spawn agents that return ready-to-review PRs.
Origin
Fiona Fung, describing how she adapted her management practice on the Claude Code and Cowork teams once code throughput increased roughly 8x.
Core principles
- 01Full visibility across repos, Slack, and metrics is a prerequisite when throughput outpaces manual tracking
- 02Use the agent not just to ship, but to generate themes and start human conversations about impact
- 03Make new mistakes — aim to learn, not to make zero mistakes, which just means being too cautious
- 04Routines raise the abstraction: instead of writing prompts, you write something that generates the prompts and PRs for you
How to run it
- 1
Enlist a standing session everywhere
Set up a persistent agent session across all repos with access to Slack channels and the metrics you track for full visibility.
- 2
Run periodic look-backs
Regularly use the session to review focus areas, what shipped, how it did in market, feedback, and any incidents.
Pro tip Ask it to generate a theme across incidents to find quality hot spots and good areas for investment.
- 3
Convert output into conversations
Use the agent's insights to have grounded conversations with the people you support about impact and bugs, not just to auto-ship fixes.
Pro tip Normalize 'make new mistakes' so these conversations are about learning, not blame.
- 4
Automate the ritual with routines
Turn recurring manual rituals into scheduled routines that summarize feedback channels and return PRs you can review when you wake up.
Pro tip As verification gets strong enough, grant a routine more autonomy to just go for it.
Watch out Watch the context-switching load — many async agents running increases the cost of catching up.
In the wild
Fiona used to start each morning with coffee reading feedback channels for gaps she could help with; she replaced it with a routine that summarizes the themes and even drafts polish-fix PRs so she wakes up to reviewable work.
→ A manual daily ritual became an automated one that spawns agents and returns ready-to-review PRs, letting her stay on top of 8x throughput.
Common mistakes
Using the agent only to generate PRs
Limiting the session to bug fixes wastes its biggest management value: surfacing cross-cutting themes and enabling impact conversations with the team.
Chasing zero mistakes
Aiming to make no mistakes signals you're moving too slowly or being too cautious; the goal is to make new mistakes and keep learning.
Is it for you?
Best for
Engineering managers whose teams ship far more than they can track manually and who want visibility plus automated rituals
Not ideal for
Small teams or low-throughput contexts where a simple manual list of what shipped still suffices
From the transcript
“I actually have a cloud code remote session that I enlist in all of our repos. And so, this way I have full visibility into…”
“I actually have these sessions to enable me to have conversations with folks that I support.”
“okay to make mistakes, just make new ones so that we're always learning cuz if you aim to make zero mistakes, like that probably means…”
“we we launched routines. And that's also completely changed. Like now I just have a routine that automates all this for me”
From the episode
What happens after coding is solved?
Fiona Fung (Manager of the Claude Code and Cowork Teams)