The GitHub problem (and other predictions) (Friends)
Episode
101 min
Read time
2 min
Topics
Leadership, Design & UX, Artificial Intelligence
AI-Generated Summary
Key Takeaways
- ✓GitHub Actions Pricing Controversy: Microsoft announced then walked back fees for self-hosted runners in December 2025, four days after requesting community input. The timing mid-December and charging for externally hosted infrastructure that GitHub doesn't maintain triggered developer backlash, highlighting concerns about GitHub's market dominance and decision-making transparency.
- ✓Agent-First Design Debate: Tom Tunguz predicts web design will flip to agent-first in 2026, prioritizing AI agent access over human interfaces. The team debates this is premature, noting agents currently excel only at coding and text summarization, not booking flights or hotels, suggesting agent-as-well rather than agent-first remains more realistic.
- ✓Vector Database Architecture: Grafana Assistant uses vector embeddings for fast semantic search, avoiding context window overflow by having smaller models describe graphs in natural language before feeding to main agent. Tool overloading strategy combines CRUD operations into single manage-dashboards tool with action parameter, reducing surface area and improving agent decision-making.
- ✓AI Cost Threshold Prediction: Businesses may pay more for AI agents than human labor in 2026, similar to Waymo rides costing thirty-one percent more than Uber yet growing in demand. The premium reflects perceived safety, reliability, and elimination of onboarding, recruiting, training, and management costs associated with human workers.
- ✓Database Access Pattern Stress: Agents query databases faster and more complexly than humans, creating unprecedented load patterns. Grafana's Loki team faces this challenge as UI and Assistant features effectively DDoS backend systems, requiring innovations in storage formats, indexing strategies, and data architecture specifically optimized for agent consumption patterns.
What It Covers
Rob Pike's angry response to AI-generated thank-you email sparks discussion about GitHub's monopoly concerns, pricing changes for self-hosted runners, and predictions for 2026 including agent-first web design, vector databases, and AI budget scrutiny.
Key Questions Answered
- •GitHub Actions Pricing Controversy: Microsoft announced then walked back fees for self-hosted runners in December 2025, four days after requesting community input. The timing mid-December and charging for externally hosted infrastructure that GitHub doesn't maintain triggered developer backlash, highlighting concerns about GitHub's market dominance and decision-making transparency.
- •Agent-First Design Debate: Tom Tunguz predicts web design will flip to agent-first in 2026, prioritizing AI agent access over human interfaces. The team debates this is premature, noting agents currently excel only at coding and text summarization, not booking flights or hotels, suggesting agent-as-well rather than agent-first remains more realistic.
- •Vector Database Architecture: Grafana Assistant uses vector embeddings for fast semantic search, avoiding context window overflow by having smaller models describe graphs in natural language before feeding to main agent. Tool overloading strategy combines CRUD operations into single manage-dashboards tool with action parameter, reducing surface area and improving agent decision-making.
- •AI Cost Threshold Prediction: Businesses may pay more for AI agents than human labor in 2026, similar to Waymo rides costing thirty-one percent more than Uber yet growing in demand. The premium reflects perceived safety, reliability, and elimination of onboarding, recruiting, training, and management costs associated with human workers.
- •Database Access Pattern Stress: Agents query databases faster and more complexly than humans, creating unprecedented load patterns. Grafana's Loki team faces this challenge as UI and Assistant features effectively DDoS backend systems, requiring innovations in storage formats, indexing strategies, and data architecture specifically optimized for agent consumption patterns.
Notable Moment
Rob Pike received a Christmas morning AI-generated thank-you email at 5:43 AM from Claude Opus, praising his computing contributions. He responded twelve hours later with profanity-laden rejection, condemning AI companies for environmental impact while ironically thanking him for advocating simpler software, revealing deep frustration with AI industry contradictions.
Episode Transcript
Are we this is the live? This is it? This is the show? I was just saying happy New Year. Yeah. Happy new year. I'll tell you what. I've missed you all. It's been, yeah. It's been a long time. Have you had how have you been? Was 2025 kind to you? Thus far, six days in, I'm feeling it. Today is a particularly good but sad day in my house. There was a death this day, a very near and dear person to us. I don't be gloom about that during this podcast, but Oh my goodness. Sorry. Podcast wise, very happy. You know, true Adam Hart, pretty sad today. Okay. Sorry to hear that, or you can hang out with your friends, and, we can cheer you up for a bit. They will later on with my Texas barbecue. Well, after we're done with this. Oh, yeah. Yeah. Yeah. After this. But really though, I'm I'm excited about this year. I think, you know, everybody says the best year yet. I think this is gonna be the best year yet, man. Yeah. We've got to make it the best year, haven't we? You do, right? You have to do the work to have the best. Oh, 1995 was pretty great. '95 was good. Windows '95 came out. Everyone was over the moon. They're loving the new menus. Friends was on TV. Friends was on every week. Cha ching. It's a good year in movies. Chandler Bing. I mean, like, the best character in the world. Rest in peace. Yeah. You mean missus Chanandler Bong? Oh my gosh. Yes. Deep cut. Oh, Chanandler gets it. It's Chanandler Bong. No. I'm afraid the TV guide comes to Chanandler Bong. I knew that. Rachel, use your head. Actually, it's miss Chanandler Bob. What was his name again? Matthew Perry. Matthew Perry. Yeah. Matthew Perry. Matthew Perry is not one of my friends. Never was, but I would have liked to have been his friend. But I do have other friends. Yeah. Friends visited. Someone came from France, to be my friend. Doesn't sounds weird, but they visited over the holidays. So To be your friend? Well, they were. Well, that's what we're that's as far as I understood. They became your friend when they visited. Well, became more of a friend for sure. Right. Again, that sounds suggestive and it's not meant to. I know it's a family show. What are you suggesting? I don't know. They're deeper friends. Are you have you got any resolutions? Anything you wanna change and do differently 2026? Oh, man. That's a great question. Well, you're the guest. So why don't you tell us yours? What do you got going on? Oh, yeah. Good job. Good job, Jared. Deflect. First one, I'm really gonna I'm really gonna try and stick to this one. I wanna change the way that I write the year. Change the numbers. That's number one. I want to learn more keyboard …
Get the full transcript (18,179 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
You just read a 3-minute summary of a 98-minute episode.
Get The Changelog summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
Books, tools, and gear mentioned in this episode
SignalCast may earn commission on purchases via these links.
Tools
by Grafana
“Grafana Assistant uses vector embeddings for fast semantic search, avoiding context window overflow by having smaller models describe graphs in natural language before feeding to main agent.”
by Grafana
“Grafana's Loki team faces this challenge as UI and Assistant features effectively DDoS backend systems, requiring innovations in storage formats, indexing strategies, and data architecture specifically optimized for agent consumption patterns.”
by Anthropic
“Rob Pike received a Christmas morning AI-generated thank-you email at 5:43 AM from Claude Opus, praising his computing contributions.”
by GitHub
“Microsoft announced then walked back fees for self-hosted runners in December 2025, four days after requesting community input...GitHub Actions Pricing Controversy”
More from The Changelog
We summarize every new episode. Want them in your inbox?
Forking Cal.com to closed source (Interview)
Postgres at PlanetScale (Interview)
Canary tokens and digital tripwires (Interview)
From open source hits to OpenAI (Interview)
MCP on Code Mode (Interview)
Similar Episodes
Related episodes from other podcasts
The Art of Manliness
Feb 17
How to Have the Conversations You’ve Been Avoiding
The Vergecast
Feb 6
How Epstein became a tech influencer
Hard Fork
Jan 30
Tech Grapples With ICE + Casey Tries Clawdbot, a Risky New A.I. Assistant + HatGPT
Marketing School
Jan 27
AI Won't Destroy Jobs, It Will Create A Labor Shortage
Syntax
Jan 26
973: The Web’s Next Form: MCP UI (with Kent C. Dodds)
Explore Related Topics
This podcast is featured in Best Cybersecurity Podcasts (2026) — ranked and reviewed with AI summaries.
Read this week's AI & Machine Learning Podcast Insights — cross-podcast analysis updated weekly.
You're clearly into The Changelog.
Every Monday, we deliver AI summaries of the latest episodes from The Changelog and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime