Taylor Dolezal

@DynamicWebPaige Definitely want to check this out 👀 and that WiFi notice is fire 😂

📚 Bell's Books in Palo Alto is my perpetual happy place. Their science fiction and mathematics (and physics!) sections always have spill-off from retired professors' book collections, there's nothing like it:

@charlieholtz @michael_chomsky Can't wait for this 🥳

@michael_chomsky mobile app almost ready! we can get you on the alpha next week

what’s the best app right now for connecting all my gh repos at once, connecting my anthropic and codex subs, and then chatting with a single agent on my iphone?? please I need this bad!!

When @github used @GitHubCopilot for code review with better tooling (for agents), the quality of reviews got worse. Rewriting your instructions to read *like* a reviewer (e.g., ask, narrow, read, decide) cuts review costs by ~20% while maintaining the same quality. Spending time on workflow design can help you far more than adopting the latest frontier model. github.blog/ai-and-ml/gith…

We ran internal benchmarks comparing @OpenAI's GPT 5.6 Sol vs @AnthropicAI Claude Fable 5 GPT 5.6 Sol performs on-par with Fable at a lower cost, but how it does this is fascinating. GPT 5.6 spends MORE time and tokens gathering context and less on formal planning before executing. This a stark contrast to previous GPT Codex models that would spend significantly less time gather context compared to Claude models. It highlights the importance and impact of context gathering for coding agents. Better context, better outputs. It seems like @OpenAI may have finally found out Claude's secret sauce.

I loved the Making of Claude Code story from @anthropic! Wild to think that internal projects like Claude Code or even Kubernetes can become such a paradigm shift (especially when they're built to solve your specific workflow challenges). anthropic.com/features/makin…

Come join us for some LLM Wiki hot takes and deep dives (live now) with @devstein64, @hwchase17, @BraceSproul, and @jeffreyhuber! events.langchain.com/webinar/llm-wi…

LLMs make writing code easier, but reviewing it is as hard as ever. Most of us now spend more time reviewing an agent's output than a teammate's. When did you last update your review process? seangoedecke.com/good-code-revi…

We need something like Makefile but for software verification: each rule specifies a file/glob that, when changed, runs linters, custom AST checks, and coding agents with custom prompts for adversarial reviews. Then you can plug it into your CI and into your agentic loops.

CodeWiki auto-generates whole-repo docs, including diagrams, and beats DeepWiki (68.79% vs 64.06% quality scores), though it's a bit shakier with system code. Do you trust generated docs for your projects? arxiv.org/abs/2510.24428

ETH Zurich tested AGENTS.md on GitHub Issues and found that auto-generated context files made coding agents ~3% worse while costing ~20% more. How are you keeping your knowledge and context fresh today? arxiv.org/abs/2602.11988

llm wikis are a glimpse of the future of what agent memory looks like this blog i wrote resonated with a lot of folks will be discussing this (as well as some updates ive made to my beliefs since this!) this thursday with @BraceSproul @devstein64 events.langchain.com/webinar/llm-wi…

🤫

Come join us on Thursday! It'll be a great session you'll want to add to your MEMORY.md 😄

@devstein64 All write, all write, all write ✍️

✍️

@charlieholtz Can’t wait to see what drops this week 🥳

got some fun stuff coming this week :)

@brian_armstrong Would love to chat with you about how we’re optimizing context @dosu_ai with your keeping context lean point. Knowledge + context benefits from TTLs

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work. Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task. Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented. Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted. Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect. The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable. Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.

Infinite information is the same as no information. Borges figured that out in 1941, in a story about a library holding every possible book. It's the idea behind this month's Dosu Drop. ☔ go.dosu.dev/5OFp3A1

Agent fuzzing was not on my bingo card for 2026

Interesting to observe Anthropic going from the moat being the best model to building a tooling ecosystem with right integrations to common dev + non-dev workflows. If I was a CTO I’d only have a Slack integration where I can switch models *anytime*… to avoid lock-in tho

Frontier models teach smaller local models with plain markdown skills! Documentation is the sole channel through which knowledge travels. tomtunguz.com/the-pi-agent-s…