|
AI Builders Digest
Thursday, June 25, 2026
|
|
The stories today all orbit the same question: what does it actually feel like when an AI joins your team? Not as a chatbot you open in a separate tab, but as something woven into the tools you already use. The answer is arriving fast, and Anthropic is doing more of the plumbing than anyone is giving them credit for.
|
|
---
|
|
01
|
Claude just showed up in your Slack channel
|
|
|
Anthropic launched Claude Tag in beta for Enterprise and Team customers on Slack. The mechanic is simple: tag Claude in a channel, give it a task, point it at your Box files. Box CEO Aaron Levie demoed this himself, showing that the integration turns any company's document library into a knowledge base accessible mid-conversation, without switching apps or uploading files manually. Andrej Karpathy called this the third major redesign of how people interact with AI, after the chatbox era and the coding assistant era. His framing: once the engineering work is done to make Claude "just work" across tools, memory, and security, it stops being software you use and starts being a colleague you talk to.
|
Why it matters: Your Slack is already where your team makes decisions. When Claude can read the relevant contract, pull the right spec doc, and draft a response without anyone leaving the thread, the question stops being "should we use AI?" and starts being "who manages the AI's access permissions?" Your IT team is about to get very busy.
|
|
Source →
|
|
Source →
|
|
Source →
|
|
---
|
|
02
|
Peter Yang notices the org chart problem nobody wants to draw
|
|
|
Peter Yang posted a short observation with a real sting to it: interacting with AI agents is converging on managing a highly capable employee. His joke about scheduling 1-on-1s and performance reviews for Claude lands because it's not entirely a joke. The more capable the agent, the more you need to think about delegation, feedback, and accountability, which are management skills, not engineering skills.
|
Why it matters: Yesterday we covered Aaron Levie's argument that evals are the real moat. Peter Yang's point is the human side of that same coin. Someone at your company has to own the relationship with the AI agent. Right now that job is falling to whoever set it up, which is usually a developer who doesn't want to be a manager and a manager who doesn't know what the agent is doing. That gap is where things break.
|
|
Source →
|
|
---
|
|
03
|
A Chinese AI lab almost nobody tracked just became one to watch
|
|
|
Swyx flagged that Zhipu AI (the company behind the GLM model family) IPO'd in Hong Kong in January and has since released what he calls the "world's undisputed top open model," beating DeepSeek. The company is returning to San Francisco and is scheduled to appear at an upcoming AI event. Swyx noted that when he first encountered Zhipu, the GLM models had almost no Western user base. That's clearly no longer true.
|
Why it matters: If you're making model selection decisions for a product, your shortlist probably has Llama, Gemini, and maybe Mistral on it. GLM-4 or its successor should probably be on that list now. "Best open model" claims circulate constantly, but Swyx has been early on Chinese lab progress before, and this one has an IPO price as a real-world signal, not just a benchmark.
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|