RANVOO sent me their AirJet X5 electric toothbrush to try, and I really like it 👌 It looks good, has different cleaning modes, and the AirJet tech gives a nice deep clean without feeling too harsh on the gums. It gently blows air while brushing, watch the video 🫧 I’ve been using it recently, and I think I’ll keep it as my daily toothbrush.

If AI triages your alerts, pulls your playbooks, or drafts your incident responses, it's not just tooling anymore. It's part of the response. Most tabletop exercises still only check if your people follow the runbook. Nobody's checking if the AI handed them the right one. You won't find that out until a live incident does it for you.

French fintech @finaryhq needed to run a custody key ceremony, with no internal experience and a regulatory need for independent attestation. 🔑 Halborn reviewed the runbook, identified two gaps, then attested the ceremony live. 🔗 Full case study:

On race day, with many pieces of software running at once, engineers have to find what is happening and where within seconds 🏎️ Until now, the logs and metrics scattered across the factory and the track had to be examined separately to find the cause. Elastic combines them into one operational view. Logstash collects the logs into Elasticsearch, and Kibana handles analysis, so engineers can trace errors across connected applications. The post says search happens in milliseconds. AI is the next step. It can combine telemetry, past incidents, and runbooks to help identify root causes, while engineers keep control of decisions. Elastic has signed a multi-year partnership with Audi Revolut F1 Team: Official Race Partner at the 2026 United States Grand Prix, and Official Search and Observability Supplier from 2027. The specific 2027 use cases are still to be validated. How Elastic helps Audi Revolut F1 Team find what matters across engineering software #Elastic# #Observability#

TL;DR A method that rewrites successes found under several harnesses into training data a single general harness can reproduce. Training on the raw traces hurts; rewriting them lifts pass@3 from 57.0% to 74.2%. Title: Scaling Trajectories for Complex Tasks through Recursive Self-Rewrite URL: Points 🧩 Different harnesses solve different tasks; the union of three solves 34.3% more than the best single one 📝 A planner turns each success into a runbook, and an executor re-solves the task from scratch under the general harness 🔍 A critic screens every runbook for leaked answers and sends rejects back with a reason 🧹 The model then flags values that can't be derived from the task or environment and discards those traces 📈 Training data grows from 2,001 to 11,094 trajectories, and Terminal-Bench Hard rises from 39.0% to 63.0% ⚠️ Direct SFT actually degrades TB2, with dead-end loops where the model repeats the same action pattern forever 🎛 Even with no training, swapping the harness doubles TBH from 33.0% to 66.0% If you're thinking about feeding your own agent logs back into training, this makes a solid case for a rewrite-and-audit step first. #AIAgents# #SelfImprovement#

Really interesting new blog post from @openai for several reasons: 1) Shows an example of building with WebMCP, meant for when you want agents and and humans to collaborate on using a UI (like co-editing notebook cells). It's different than MCPs or APIs in that its exposed directly through the browser. Read the post for discussion of the tradeoffs. 2) They created a new kind of notebook which works with WebMCP that prioritizes meeting people where they are: you bring your own coding agent and files are just markdown. The author uses it to curate runbooks or high quality examples of how to run foundation model evals on their infrastructure. Notebooks are good for this since they require tinkering with state of long running jobs interactively while taking notes inline. And its open source ✨ Blog: P.S. this post is authored by Jeremy Lewi who isn't on X but here is his website

TL;DR: Agents fail not from a lack of model smarts but from a weak execution harness. "Harness scaling" — hardening the runtime without touching weights — reaches 95.3% on Terminal-Bench 2.1 for ~$15. Title: StateM: Reaching 95.3% Raw Accuracy, or a $15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling URL: Key points 🧠 Scale the harness, not the model — learn from failures with no retraining 📌 Durable states retain discoveries across episodes 🎯 Phase-local context surfaces only relevant info at decision points ✅ Checked transitions verify preconditions to stop cascading failures 📖 Recoverable runbooks turn failures into queryable knowledge 🕒 Versioned practices let agents update and roll back strategies 💰 GPT-5.6 hits 95.3% (445 trials, all 89 tasks), compressing a $574.68 run to ~$15 It turns every failure into an executable precondition, lifting reliability from the operational layer. #AIAgents# #TerminalBench#

# Practical and Useful Patterns with ADK ## 📚 Load Knowledge On-Demand with ADK Skills Stuffing all knowledge into your agent's instructions blows up token costs. ADK **Skills** let you load workflows and knowledge on-demand — only when relevant — saving tokens while extending capabilities! 🚀 ## 📌 Title Skills for Agents ## 🔗 URL ## 🧩 Overview A Skill is a self-contained unit of functionality that an ADK agent can use to perform a specific task. It packages instructions, references, assets, and scripts, loading them only when the agent needs them. Skills follow a 3-level architecture: - **L1 (Metadata)**: `SKILL.md` frontmatter enables skill discovery - **L2 (Instructions)**: Main instructions in the `SKILL.md` body - **L3 (Resources)**: Supporting materials in `references/`, `assets/`, `scripts/` By loading only when needed, Skills optimize context window usage and minimize token consumption. ## 🛠 How to Use **Directory structure:** ``` my_agent/ skills/ code_review/ SKILL.md # Required: metadata + instructions references/ CHECKLIST.md # Detailed review checklist assets/ template.json # Review output template scripts/ # Lint execution script ``` **Python integration:** Import `load_skill_from_dir` from `google.adk.skills` and `skill_toolset` from ` along with `pathlib`. Call `load_skill_from_dir()` with the path `pathlib.Path(__file__).parent / "skills" / "code_review"` to load the skill from its directory. Then create a `skill_toolset.SkillToolset(skills=[code_review_skill])` and pass it to the `Agent` constructor via the `tools` parameter. The agent is created with `model="gemini-flash-latest"` and `name="dev_assistant"`, gaining on-demand access to the skill through the toolset. ## 🏗 Practical Usage **Skills for a DevOps agent:** ``` skills/ code_review/ SKILL.md # "Code Review Checklist" references/ CHECKLIST.md # Detailed review criteria release/ SKILL.md # "Release Procedure" references/ PROCEDURE.md # Step-by-step release flow assets/ changelog.tmpl # CHANGELOG template incident/ SKILL.md # "Incident Response" references/ RUNBOOK.md # Incident response runbook scripts/ # Slack notification script ``` The agent loads `code_review` only when asked to review code, and `release` only when asked to deploy. No need to keep all skills loaded at all times, dramatically reducing token consumption. **Inline skills (TypeScript):** ```typescript const releaseSkill: Skill = { frontmatter: { name: 'release-procedure', description: 'Standard release procedure with checklist', }, instructions: "Follow the release checklist step by step...", resources: { references: { 'checklist.md': '1. Run tests\n2. Update version\n3. Create tag...', }, }, }; const skillTools = new SkillToolset([releaseSkill]); ``` ## 💡 Use Cases - 📋 **Code review checklists**: Load detailed review criteria only when reviews are requested - 🚀 **Release procedures**: Load step-by-step deployment guides and templates only at release time - 🔧 **Troubleshooting**: Load runbooks and notification scripts only during incidents - 📝 **Document authoring**: Load type-specific templates and guidelines per document type - 🎓 **Employee onboarding**: Load relevant knowledge base sections based on questions ## ⚠️ Caveats - Skills are experimental (Python v1.25.0+, TypeScript v0.6.1+, Go v1.2.0+) - `SKILL.md` is required; include `name` and `description` in frontmatter - Skill discovery accuracy depends on description quality — write clear descriptions - Large resource files impact token consumption; keep them minimal - Be mindful of script execution permissions and security ## ✨ Closing Skills implement "load knowledge only when needed" in an elegant way. Instead of cramming everything into instructions, modularize capabilities as Skills to keep token costs low while extending your agent's abilities infinitely. DevOps, customer support, education — the pattern applies everywhere! #ADK# #AIAgent#