Telexed

AI news through a solo-operator lens — only what changes your day8 of 829

FILTER[All][Agents & tools][Models & API][Generative media][Infra & SaaS][ASO & growth][Indie business][Idea signals][Other][★6+ high-signal]

Simon Willison ✕clear filters

Fri, May 291 dispatches

#0829
#0829Agents & tools Simon Willison2 days ago
`llm-anthropic` `0.25.1` adds `Claude Opus 4.8` and fast mode
50radar
llm-anthropicLLM CLI plugin — calls Anthropic models from `llm`
Output caps now default to each model's real maximum, not 8,192. Useful for long generations through llm; fast mode only matters if your Anthropic org has access.
- Adds claude-opus-4.8, so the llm CLI can call Anthropic's newest Opus model without custom aliases.
- New -o fast 1 flag exposes Anthropic fast mode, but only enabled organizations can use it.
- Default max_tokens now follows each model's maximum output instead of fixed 8,192, reducing silent truncation risk.
Source: simonwillison.net/2026/May/28/llm-anthropic/#atom-everytRead original →
50radar
PHOTO
FIG-8291:1

Thu, May 282 dispatches

#0828
#0828Agents & tools Simon Willison3 days ago
`SQLite` Adds `AGENTS.md` Rules for Coding Agents
50radar
A major OSS project now draws a hard line: agent-written patches are out, reproducible bug reports are welcome. Treat AGENTS.md as repo-level policy, not just agent prompt decoration.
- SQLite says it does not accept agentic code, and removed “currently” to make that stance permanent.
- Agent-generated bug reports still have a path if they include a reproducible test case. That is the useful contribution format.
- The forum was split after a flood of AI bug reports. Maintainers need intake rules before agent traffic arrives.
- Adding AGENTS.md to your own repo can steer agents toward tests, repro steps, and contribution boundaries.
Source: simonwillison.net/2026/May/27/sqlite-agents/#atom-everytRead original →
50radar
PHOTO
FIG-8281:1
#0827
#0827Agents & tools Simon Willison3 days ago
Coding agents are shifting from seat discounts to API-priced usage
50radar
Flat seats hide heavy Claude Code and Codex usage. Once teams pay API rates, agent adoption turns into a real infra cost line, not just a productivity perk.
- A $200/month personal setup mapped to $2,180.16 in estimated API tokens over 30 days; flat plans are a subsidy for heavy users.
- Anthropic Enterprise reportedly moved to $20/seat/month plus API usage within the last six months, changing team economics fast.
- Budget surprises are a demand signal. If staff keep using agents despite high bills, OpenAI and Anthropic have pricing power.
- For small teams, agent ROI needs tracking like hosting or database spend: caps, logs, and per-workflow cost checks.
Source: simonwillison.net/2026/May/27/product-market-fit/#atom-eRead original →
50radar
PHOTO
FIG-8271:1

Mon, May 252 dispatches

#0826
#0826Agents & tools Simon Willison6 days ago
`datasette-agent` `0.1a4` adds chat launch from Jump menu
50radar
datasette-agentDatasette plugin — runs database queries through agent chat
Agent chat now sits behind the / shortcut inside Datasette. Nice pattern for making data tools conversational without adding another heavy UI surface.
- Uses Datasette 1.0a30's new makeJumpSections() JavaScript plugin hook, so the agent entry point lives inside the existing Jump menu.
- Typing a request like count entries starts an agent conversation and returns a database answer; the demo result was 3300 entries.
- Available on agent.datasette.io with GitHub sign-in. Worth copying as a lightweight command-palette pattern for internal tools.
Source: simonwillison.net/2026/May/24/datasette-agent/#atom-everRead original →
50radar
PHOTO
FIG-8261:1
#0825
#0825Agents & tools Simon Willison6 days ago
AI-polished GitHub issues are making bug reports worse
40radar
Bug reports lose value when LLM rewrites replace observed facts with confident guesses. Keep reports to commands, expectations, actual output, and exact logs; faster triage beats verbose analysis.
- Useful report shape is four fields: command run, expected result, actual result, exact error or log.
- LLM rewrites often add fake minimal repros, guessed root causes, and irrelevant error classes. That slows maintainers down.
- Best workflow: use AI to shorten wording, not invent diagnosis. Preserve the human observation as the source of truth.
Source: simonwillison.net/2026/May/24/armin-ronacher/#atom-everyRead original →
40radar
PHOTO
FIG-8251:1

Fri, May 222 dispatches

#0824
#0824Agents & tools Simon Willisonlast week
`Datasette Agent` brings conversational SQL and charts to Datasette
70radar
Datasette AgentData query AI agent — SQL and charts inside Datasette
Natural-language querying now lands inside existing SQLite-backed Datasette apps. Useful for lightweight admin analytics without building a custom BI layer.
- Datasette Agent answers questions over data stored in Datasette, turning prompts into SQLite queries and responses.
- The datasette-agent-charts plugin adds chart generation, making quick visual exploration possible without a separate dashboard stack.
- The live demo runs on Gemini 3.1 Flash-Lite: cheap and fast is enough for many SQL-writing agent workloads.
Source: simonwillison.net/2026/May/21/datasette-agent/#atom-everRead original →
FIG-8241:1
70radar
FIG-8241:1
#0823
#0823Agents & tools Simon Willisonlast week
`datasette-agent-sprites` `0.1a0` runs agent commands in `Fly Sprites`
40radar
datasette-agent-spritesDatasette Agent plugin — runs commands in Fly sandboxes
Agent command execution moves into an isolated sandbox. Early alpha, but useful if you are exposing database-adjacent tools to LLM agents.
- datasette-agent-sprites is a Datasette Agent plugin for running commands inside a Fly Sprites sandbox.
- The release is 0.1a0, so treat it as experimental rather than production-ready infrastructure.
- The practical signal is sandboxed agent execution: useful for reducing blast radius when agents run shell commands.
Source: simonwillison.net/2026/May/21/datasette-agent-sprites/#aRead original →
40radar
PHOTO
FIG-8231:1

Thu, May 211 dispatches

#0822
#0822Agents & tools Simon Willisonlast week
`Gemini Spark`, Google’s hosted agent tied to Workspace apps
60radar
Gemini SparkHosted AI agent — native Google app connections
Google is packaging app-connected agents around Workspace, but much is still coming soon. Security and credential handling decide whether it becomes useful or risky.
- Gemini Spark connects natively to Gmail, Calendar, Drive, Docs, Sheets, Slides, YouTube, and Maps. That makes it closer to a work agent than a chat UI.
- FAQ says it runs on Gemini 3.5 Flash and Antigravity. The Antigravity stack spans a desktop app, CLI, SDK, and VS Code fork.
- Enterprise notes mention fresh isolated ephemeral VMs, Agent Gateway DLP, and encrypted credentials. That is the right threat model, not proof it is solved.
- Since the product is not broadly testable yet, treat it as a roadmap signal. Do not build critical automations around it until GA behavior is clear.
Source: simonwillison.net/2026/May/20/google-io/#atom-everythingRead original →
60radar
PHOTO
FIG-8221:1

`llm-anthropic` `0.25.1` adds `Claude Opus 4.8` and fast mode

`SQLite` Adds `AGENTS.md` Rules for Coding Agents

Coding agents are shifting from seat discounts to API-priced usage

`datasette-agent` `0.1a4` adds chat launch from Jump menu

AI-polished GitHub issues are making bug reports worse

`Datasette Agent` brings conversational SQL and charts to Datasette

`datasette-agent-sprites` `0.1a0` runs agent commands in `Fly Sprites`

`Gemini Spark`, Google’s hosted agent tied to Workspace apps