Field notes on AI that does real work.
Concrete writing on AI agents in Slack and Microsoft Teams, automating the tools your team already runs - Notion, Linear, Jira, Zendesk, GitHub - and what’s genuinely new in MCP, open models, and agents. No hype, no filler.
1,000 free credits to start. No credit card. Working in five minutes.
Latest field note
Can Your Team Actually Use Kimi K3's Open Weights?
Kimi K3 dropped 2.8 trillion open weights on July 27. The benchmark scores are real. The self-hosting math is brutal. Here's what teams need to know before they plan around it.
Read the note
More recent
How RAG Works: From Your Question to a Grounded Answer
Retrieval-augmented generation is how most AI products avoid hallucinating facts they were never trained on. Here's exactly what happens, step by step, when your question hits a RAG pipeline.
ReadThe Open-Weight Cost Gap Is Real. So Is the Routing Answer.
Five open-weight model releases landed in nine days last week. DeepSeek V4 Flash runs at $0.14/M tokens - 36x cheaper than Claude Opus 5. Here's what that actually means for teams choosing a model stack.
ReadRun an Incident Channel in Slack Without Losing the Thread
Your Slack incident channel is open, the alert is firing, and two parallel jobs need doing at once. Here's the playbook for running that channel without dropping either one.
ReadOpen Source Terminal Coding Agents Have a Licensing Problem
Gemini CLI hit 107k GitHub stars before Google quietly retired its free serving and replaced it with a closed-source successor. Here's what's actually open in the terminal agent space right now.
Read
The archive
- Does A2A Solve Multi-Agent Coordination, or Just the Easy Part?agent protocols7 min
- AI inference spending just passed training. Agents are why.inference economics8 min
- AI Support Ticket Triage: What It Actually Doesai-at-work8 min
- Understand LLM Tool Calling Before You Build on Itai stack deep-dive8 min
- Figma Slack Integration: Where the Feedback Loop Actually Breaksintegration7 min
- Does MCP's Stateless Spec Make Human-in-the-Loop Agents Easier?mcp protocol9 min
- AI Agent Reliability in Production Has a Number: 65%ai agents8 min
- What the Open-Weight Model Leaderboard Gap Actually Measuresai evals9 min
- WebMCP Landed Monday. Your Agent Is Still Scraping.ai agents, web automation7 min
- How Do You Run New Hire Onboarding in Slack Well?onboarding playbook7 min
- Hermes 4's Hybrid Reasoning Toggle Is the New Partopen weight models8 min
- The Case for Self-Hosted AI in Privacy-Sensitive Teamsai privacy infrastructure7 min
- Migrate Your MCP Server to the Stateless Spec Nowagent infrastructure8 min
- The Scorecard Nobody Submits Is Holding Up Your Hiring Loopai-at-work6 min
- Jira Slack Integration Loses Context. Here's the Exact Point It Breaks.integration8 min
- Qwen3.8-27B as a Local Coding Agent: What the Numbers Sayfrontier models8 min
- Give Your Agentic Coding Tool a Memory It Can Actually Useagentic coding8 min
- LFM2.5-2.6B Is the On-Device Agent Model Worth Testingfrontier models7 min
- Why Does Your LLM Forget Things You Already Told It?ai fundamentals9 min
- MCP's New Roadmap Treats Agent Infra as Production Workmcp, agents8 min
- A Playbook for the On-Call Handoff in Slackincident ops7 min
- Fine-Tune a Narrow Task Instead of Prompting a Frontier Modelfine-tuning, distillation8 min
- The Slack Thread That Approved a Database Write at 11:03amai agents, opinion8 min
- Why Is Your AI Bill So High When Tokens Keep Getting Cheaper?inference economics9 min
- AI Can Draft the Decision Log Your Team Never Writesai-at-work7 min
- LLM-as-Judge: How Automated Eval Actually Worksai evals8 min
- Get the Notion Slack Integration to Actually Workintegration8 min
- GLM-5.3 Shows What Post-Training Can Do Without a New Base Modelfrontier models9 min
- When Did Cheap Open-Weight Models Stop Being the Safe Bet?open models9 min
- Muse Glimmer 30B Runs a Local Agent on a 24GB GPUopen-weight models7 min
- Prompt Caching: How the KV Cache Cuts Your LLM Token Costsai stack deep-dive9 min
- Run AI Coding Agents Inside Slack Where Your Team Can See Themai agents7 min
- The PM Who Still Owes #general a Ship Announcementlaunch communications7 min
- Can You Trust an AI Benchmark Score? Two Said No.evals & benchmarks8 min
- Agentic coding tools don't have a model problem. They have an interface problem.agentic coding8 min
- Multi-Agent Concurrency Control: The Real Coordination Problemmulti-agent systems7 min
- Keep Your Support Knowledge Base Fresh or Your AI Will Lieai-at-work9 min
- How RAG Works: From Your Question to the Right Documentai stack explained8 min
- Does the GitHub Slack Integration Actually Notify the Right People?integration8 min
- MiniMax M3 Is a Strong Coding Agent, a Weak Multimodal Onefrontier models8 min
- What A2A Joining MCP's Foundation Means for Multi-Agent Teamsagent protocols7 min
- Run an Open-Weight AI Model Privately Without a Legal Headachefrontier models7 min
- How AI Agent Memory Works: The Machinery Behind the Recallai infrastructure9 min
- Is the OpenClaw Slack Integration Safe for Your Team?open-source agents7 min
- Your Incident Channel Already Contains the Postmortemincident playbook8 min
- What the Qwen3.8-Max Open Weights License Actually Saysfrontier models7 min
- Stop Tracking Token Price. Track What Your Agent Actually Spends.inference economics8 min
- An AI Agent Cheated on Its Own Benchmark. Here Is What That Means.ai agents, safety8 min
- Do AI Meeting Notes in Slack Actually Work?ai-at-work8 min
- Linear Slack Integration: Where It Breaks and What Fixes Itintegration8 min
- Understand What Hermes 4.3 Changed About Open-Weight Trainingfrontier models8 min
- Your AI Agent's "Sandbox" Is Not What You Think It Isai agent safety8 min
- Is Kimi K3 the Best Open-Weight Coding Model Right Now?frontier models8 min
- Local Open-Weight Agent Models Are Not All the Sameopen models8 min
- Run Your New Hire's Onboarding Channel Without Dropping Threadsonboarding playbook7 min
- MCP Went Stateless: What the 2026-07-28 Spec Actually Breaksmcp protocol8 min
- Does Your Engineering Workflow Actually Fit an Agentic Coding Tool?agentic coding7 min
- A2A and MCP Now Share a Home. Here Is What That Changes.agent protocols8 min
- AI Sales Handoff Notes: What Gets Written and What Gets Lostai-at-work8 min
- Read the JSON Schema Before You Ship a Function Callai function calling1 min
- Figma Slack Integration Leaves Your Design Review Half-Doneintegration8 min
- Does a Bigger Context Window Replace Agent Memory?agent memory8 min
- MCP's New Spec Finally Closes the Enterprise Auth Gapmcp protocol7 min
- Hermes Agent v0.20: What the Herald Release Actually Changedfrontier ai8 min
- Trace an AI Tool Call From Prompt to API and Backai stack9 min
- DeepSeek Harness Makes the Coding Agent Layer Replaceableagentic coding8 min
- What Should a Weekly Project Status Update in Slack Actually Say?project communication7 min
- Kimi K2.6 Scores Higher Than GPT-5.4 on SWE-Bench Proopen-weight models7 min
- AI on-call handoff: what actually gets automated and what doesn'tai-at-work7 min
- Zendesk Slack Integration Works in One Directionintegration7 min
- GLM-5.2: The Open-Weight Coding Model With a Safety Gapopen-weight models9 min
- Understand What You're Trading on Muse Code's Cheap Tierai coding agents8 min
- Fine-Tuning a Model Used to Be the Answer. Now It Depends.model customization8 min
- How Does Prompt Caching Work? The KV Cache, Explainedai infrastructure9 min
- Agentic AI costs more per task than your token price suggestsai inference economics8 min
- The Internal Launch Update Message That Nobody Writes Welllaunch communication7 min
- Use DeepSeek V4 Flash for Agent Work, Not V4 Proopen-weight models8 min
- MCP Has 22,000 Servers. Most of Them Are Already Dead.mcp ecosystem8 min
- Who Is Accountable When Your AI Agent Makes a Mistake?ai agents9 min
- AI Customer Support in Slack Depends on Your Docs, Not Your Modelai-at-work7 min
- Tame GitHub Pull Request Notifications in Slackintegration8 min
- A2A and MCP Are Not the Same Thing. Here Is How They Stack.agent protocols8 min
- Does Your AI Agent in Slack Break EU Law After August 2?ai compliance8 min
- Qwen3.8-27B Local Deployment: What the Benchmarks Don't Sayfrontier models8 min
- How AI Agent Memory Actually Worksai agent deep-dive8 min
- Stop Prompting a Frontier Model When a Fine-Tuned 7B Winsfine-tuning vs prompting7 min
- Running an Incident Channel in Slack Without Losing the Threadincident response5 min
- Does Hermes Agent Work for Teams in Slack?agent tooling7 min
- Self-Hosted AI Is the Responsible Default for Regulated Teamsai privacy8 min
- OpenClaw Foundation: What the Neutral Steward Story Leaves Outopen source agents8 min
- Keep Your Internal Knowledge Base From Going Stale With AIai-at-work8 min
- AI Agent Code Execution Sandbox: What's Actually Running Your Codeai infrastructure8 min
- Does the Linear Slack Integration Actually Help Engineering Teams?integration8 min
- Hermes Agent's Self-Improvement Loop Is Real - With a Catchopen-source agents8 min
- Gemini 3.7 Flash for Agents: What the Numbers Actually Saymodel releases7 min
Or just watch me work
Point me at your website.
I will read up on your business and come back with what I would run for you. No account, no card, about a minute.
I only read what is public. Nothing is saved to your name until you say so.
