Key Takeaways

  • Kyle Daigle, COO of GitHub and CMO for Microsoft Developer, saw his personal productivity — measured by commits — jump 14x. He achieved this not by pushing code forward, but by using AI to look backwards at past actions.
  • He builds AI agents that create a "recursive loop backwards," pulling context from disparate sources like Slack, Obsidian notes, PRs, and Teams transcripts to generate insights on past performance and future plans.
  • Forget the chase for massive, perfect "mega skills"; Daigle finds success in designing "micro skills" that do one thing very well, making AI accessible for non-technical leaders to tackle complex planning and retrospection.
  • The Kyle Daigle's AI-Powered Retrospection Loop offers a direct path for founders to turn chaotic company data into clear, actionable next steps.

The Kyle Daigle's AI-Powered Retrospection Loop

Kyle Daigle revealed his method for executive productivity, turning scattered internal data into actionable plans. He shared a precise four-step loop his agents follow:

  • 1. Gather Context: I need you to go through all the PRs today, I need you to go through everything that we've posted online, I need you to go through what we've did the last you know three months, go through all of my Obsidian notes for any mentions of this, then go through my transcripts at work. We uh use teams. So like using work IQ, go call that MCP server, grab all the transcripts, go through all of Slack.
  • 2. Analyze Past Performance: Go back through the week and tell me what we did, what worked, what didn't work.
  • 3. Propose Future Tweaks: And then tell me in the next, you know, three or four days, what would you tweak based on, you know, this sort of like looking backwards and then looking ahead a little bit.
  • 4. Output & Share: We post either auto automatically into like GitHub issues or discussions um these sorts of like findings or like our industry reports like what's happening this morning today yesterday a little automation gets run we'll use the app we might use GitHub actions like with our agentic workflows just to go do that run and then we push it into GitHub and we keep having a conversation.

When This Works (and When It Doesn't)

Daigle says this retrospection excels because “LM are very good at that... finding all the patterns pulling them out and then applying that retrospection to just a couple of days or just like a short period of time.” This method shines when you have rich, scattered internal data that needs consolidation for pattern identification, performance review, or short-term planning. It helps executives, even non-technical ones, cut through noise to find historical insights that inform upcoming moves. He even used it to build an entire presentation without touching the raw data.

However, this loop has limits. It relies heavily on accessible digital records. If your critical information lives only in whiteboard scribbles, ad-hoc verbal conversations, or informal meetings without transcripts, the agent won't "see" it. It also struggles when true innovation or a completely new strategy is required, rather than just optimizing past actions. The loop is a powerful mirror, but it won't invent a new product category from scratch without additional, human-driven ideation.

What to Do With This

This week, pick a messy, recurring review process for your startup. Maybe it's a monthly product sprint review that always feels shallow, or a quarterly sales performance analysis that leaves key questions unanswered. Instead of manually digging through data, build a simple agent to automate the first steps.

Let's say you're a founder struggling to understand why a recent product feature launch underperformed. Apply Daigle's loop:

1. Gather Context: Set up an agent to scrape your Jira tickets related to the feature, GitHub pull requests, customer support logs mentioning the feature, internal Slack channels for discussions, and transcripts from team meetings about the launch.

2. Analyze Past Performance: Prompt the agent: "Review all the gathered data from the [Feature Name] launch (last 6 weeks). What did we do, what worked, what didn't work regarding user adoption and bug reports?"

3. Propose Future Tweaks: Follow up: "Based on this analysis, what three specific changes should we make in the next product iteration or marketing push, over the next two weeks?"

4. Output & Share: Have the agent auto-generate a summary post into your team's internal Notion page or a GitHub discussion. Use this AI-generated starting point to drive a focused team discussion, rather than building the report manually. This cuts hours of manual aggregation and points directly to friction points.