A collection of tools and experiments built with tldraw offline, the local whiteboard for you and your agents. Download and remix them with a coding agent.
tldraw.dev/blog/buildi... 
"The blue rectangle. Top left. No, the other one." About giving an agent a way to point instead of describing, by Tim Gilboy, Head of Product at tldraw.
tldraw.dev/blog/agents... About giving an AI agent a way to actually show its thinking instead of just describing it, by Steve Ruiz.
tldraw.dev/blog/text-i... We spend a lot of time building on the canvas. We write a lot about it, too. In the thread below, a few recent pieces from the archives on AI, agents, and spatial software: 
Step 5 Preview is now live in Command Code.
Available across all plans and API. Grok 4.7 is now live in Command Code.
40% for the next 7 days
$35 credits (3.5x usage) in GOAT
Available on GOAT, Pro, Max, and via API.
🐐 MiMo V2.6 Pro & Flash are live in Command Code 🐐
Available on all plans.
Command Code has the best usage in market for both models.
E.g. in $10 GOAT plan:
• MiMo-V2.6 Flash $67 (6.7x usage) for 72hrs
• MiMo-V2.6 Pro $20 (2x usage)
🐐 
How do you build the fastest text search in Postgres? Ask the two engineers who did it. TIN TALK: live Q&A with Eric and Patrick on our Discord, Monday 11am PT. Join to ask your questions: pscale.link/community
A Postgres replica can finish replaying WAL and still not be safe to promote. If it doesn't have your logical replication slots, your change stream breaks on cutover. See how PlanetScale solves this:
Congratulations to @SpaceXAI on the @grok 4.7 drop, we've all been waiting! It's very fast, and the intelligence is quite good. Running multiple cloud agents with it now Spin it up in cloud agents with `railway code --grok`
The build is only as good as the plan. JJ Englert shows how creating a detailed implementation plan first helps AI agents identify dependencies, parallelize work, and build more efficiently.
Not your mother's CSS As with any time spent fiddling with CSS, I had a good time writing this!
We're extending $60 usage on DeepSeek V4.1 Flash in Command Code for another 8 days!! 🐐
6x usage on $10/mo GOAT
$60 usage. 154K requests. 7.8B tokens.
GOAT plan is the best AI coding plan.
Available in all plans and API. Our engineering and design team has been testing 26+ side-by-side comparisons across frontier and open models. All runs are public and open source. Benchmark for this demo here:
github.com/CommandCode... Ran FlappyBench on Qwen3.8-Omni-Flash, DeepSeek V4.1-Flash & Gemini 3.8 Flash with the same /design prompt.
🔹 Qwen3.8-Omni-Flash: 9/10 · $0.013 · smooth gameplay & cheaper
🔹 DSV4.1-Flash: 9/10 · $0.0089 one-shot with the lowest cost
🔹 Gemini-3.8-Flash: 5/10 · $0.205 · hard-coded gameplay 
Hello
GLM-5.3 FlashX is now live in Command Code.
· High speed variant of GLM-5.3 Flash
· ~200 TPS with 1M context window
· Multimodal (text, image, video)
Available across all plans and API. Your agents can finally move out of localhost
Changelog #0308
• Railway Sandboxes are generally available
• Use your ChatGPT subscription with Railway Agent
• MySQL PITR
• Fixes and improvements
railway.com/changelog/2026-09-18-railway-sandboxes Let's take a closer look at how Neki is built. Starting with real Postgres, we unpack the architecture one component at a time:
We wanted more from full text search in Postgres, so we built TIN. It gives you fuzzy matching, NEAR, wildcards, boolean grouping, and BM25 scoring. Plus, it's really fast. See it in action in our latest video:
Natural language is becoming part of the developer workflow. JJ Englert and KP discuss how AI is reducing the amount of time spent on technical setup so builders can focus more on creating and iterating.
fond memories of that place (before they moved)
Download the public beta: delta.dev/download Build a feature with an agent, have another agent review it, then bring in your teammates to approve the change. Agents and teammates, working together in Delta.
