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.
Latest field note
How Does LLM Tool Calling Actually Work?
LLMs don't execute code - they write a JSON note describing what should happen. Here's the full loop: schema, token prediction, constrained decoding, and your code doing the real work.
Read the note
More recent
Figma Slack Integration Delivers Notifications, Not Decisions
The Figma Slack integration routes comment pings to your channels, but the design decisions those comments contain still evaporate. Here's where the workflow actually breaks-and what to do about it.
ReadA2A Protocol's v1.0 Release and What It Actually Fixed
Google's A2A protocol hit v1.0 in April 2026 after a year of production use. Here's what changed, what the signed Agent Card solves, and where the spec still leaves you on your own.
ReadDeploy Your Enterprise AI Agent Like Cisco, Not Like a Pilot
Cisco is rolling out a personal AI agent to all 90,000 employees this month. The architecture - on-premises, model-routing, not always frontier - is the real lesson for teams planning internal agent deployments.
ReadHermes Open-Weight Model Picks Up Where Llama Left Off
Nous Research's Hermes 4 and 4.3 bring hybrid reasoning and decentralized training to open-weight models. Here's what actually changed, and what it means for teams running agentic pipelines.
Read
The archive
- How Does RAG Work, From Document to Grounded Answerai stack deep-dive5 min
- OpenClaw's Mobile App Gets the Architecture Rightopen source agents5 min
- A Slack Incident Channel Runbook That Actually Gets Runincident management5 min
- Run GLM-5.2 as Your Coding Agent Before Trusting the Benchmarksfrontier models6 min
- The Procurement Risk Hidden in Every AI Benchmark Scoreai evaluation6 min
- Can a Voice Agent Replace Your Three-API Call Stack?voice agents5 min
- On-Call Rotation Burnout Is an Alerting Problem, Not a Rota Problemai-at-work5 min
- How LLM Evals Work, From Test Case to Gradeai evals6 min
- Get Your Jira Slack Integration Past the Notification Wallintegration5 min
- What the SWE-bench Collapse Tells You About AI Leaderboardsfrontier evals6 min
- Does a Coding Agent Actually Change How Engineers Work?agentic coding6 min
- Model Distillation Cuts Inference Cost When the Task Is Lockedfrontier models5 min
- The LLM Context Window, From Token Count to Real Behaviorai fundamentals6 min
- Pick an Open-Weight Model Before You Sign Another API Contractopen models7 min
- Writing an Incident Postmortem in Slack Without Losing the Threadincident response6 min
- What Makes OpenCode the Most-Starred Coding Agent?open source agents5 min
- Most MCP Servers Are Not Safe to Connect to an Agentmcp security6 min
- The Persistent Memory Problem China Just Made Visibleai agents6 min
- Get Your AI Meeting Notes to Actually Drive Actionai-at-work6 min
- Your AI Forgot What You Said an Hour Ago. Here's Why.ai infrastructure6 min
- Can Notion AI Actually Answer Questions in Slack?integration6 min
- Poolside Laguna XS 2.1 Runs Agentic Coding on One GPUopen-weight models5 min
- AI Agents in Production Are Harder Than the Demos Suggestai agents6 min
- Read AI Benchmark Scores Like the GPT-5.6 Sol Evalfrontier evals6 min
- How AI Agent Memory Works, From Query to Retrieved Factdeep-dive5 min
- Should You Fine-Tune or Write a Better Prompt?ai models, agents6 min
- Product Launch Update Slack Channel: A Field Playbooklaunch playbook6 min
- AI Inference Cost: Why Cheaper Tokens Raise Your Billinference economics7 min
- Stop Treating Self-Hosted AI as the Default Privacy Answerai privacy6 min
- Claude Sonnet 5 Raises the Floor on Agentic Codingagentic coding6 min
- Can AI Actually Keep a Decision Log Without You?ai-at-work5 min
- Prompt Caching Cuts Your LLM Input Bill by Up to 90%ai infrastructure5 min
- The Read-Only Wall in Your HubSpot Slack Integrationintegration6 min
- Pick Your Open-Source Coding Agent Before the Ecosystem Shifts Againfrontier tools5 min
- GitHub Copilot Usage-Based Billing Changed How Agents Costagentic coding5 min
- Does the MCP 2026-07-28 Release Candidate Break Your Server?frontier protocol8 min
- How LLM Tool Calling Works, From Token to Executionai stack deep-dive5 min
- AI Model Export Control Risk Is Now an Ops Problemai governance5 min
- Run New Hire Onboarding in Slack Without Dropping the Threadonboarding playbook5 min
- Agent-to-Agent Protocol: What A2A Actually Doesfrontier protocols6 min
- Is Chat the Wrong Interface for AI Agents?ai agents6 min
- Hermes Agent Reaches Slack and Teams With Persistent Memoryopen-source agents6 min
- AI Sales Handoffs: What Actually Changes for AE-to-CSM Handoffsai-at-work6 min
- Understand AI Agent Sandboxing Before Your Agent Writes Codeai agents6 min
- When a Figma Comment Lands in Slack With No Contextintegration6 min
- Can DeepSeek V4 Flash Replace a Frontier Model in Your Agent Pipeline?frontier models5 min
- Claude Sonnet 5 Is Built for Agents, Not Just Better Chatmodel releases6 min
- GLM 5.2: What the Open-Weight Leader Actually Does Differentlyfrontier models6 min
- Teach Your AI Agent to Remember Across Sessionsagent memory7 min
- When Your AI Agent Asked X What the Market Thinksmcp ecosystem5 min
- What Should an On-Call Handoff in Slack Actually Contain?incident ops6 min
- MiniMax M3 Open-Weight Model: What the Numbers Actually Meanfrontier models5 min
- The SaaS Moat That Actually Survives AI Agentsai agents, saas7 min
- Publish Your ai-catalog.json Before Agents Hard-Code Around Youagent infrastructure4 min
- The Slack DM a CSM Sends the Morning After a Deal Closesai-at-work6 min
- What Do AI Evals Actually Measure?ai evals7 min
- Jira Slack Integration: What Works and Where It Breaksintegration6 min
- Kimi K2.6: The Open-Weight Model Built for Long Agent Runsfrontier models5 min
- Trim Your MCP Server List Before It Trims Your Agentmcp security6 min
- Why Cheaper Tokens Made Your AI Bill Worseinference economics6 min
- How Does an LLM Context Window Actually Work?ai fundamentals6 min
- Self-Hosted AI Agent Security: What OpenClaw Exposedai agents6 min
- AI Incident Postmortem Playbook for Slack Channelsincident response6 min
- Run GLM 5.2 Before You Renew That Claude Contractfrontier models6 min
- AI Agents Per-Seat Pricing: The Model Is Already Brokenai economics6 min
- What Happens When Your AI Model Gets Switched Off?ai reliability6 min
- AI Support Ticket Triage Cuts Misrouting by Halfai-at-work5 min
- How RAG Works: From Your Question to Retrieved Textai stack deep-dive6 min
- Keep Your Notion Workspace From Becoming a Ghost Townnotion workflow6 min
- Open-Weight vs Closed Frontier Models: What the Bill Actually Showsfrontier models6 min
- What Does GPT-5.6 Sol's Ultra Mode Mean for Your Agent Stack?ai agents5 min
- MCP 2026-07-28 Release Candidate Has Breaking Changesprotocol releases6 min
- Prompt Caching in LLMs: What It Does to Your API Billai infrastructure7 min
- Give Your AI Agent Only the Permissions It Can Justifyagentic ai security6 min
- The Slack Message Nobody Writes Before the Feature Shipslaunch playbook5 min
- What Does the MCP Stateless Spec Actually Change for Your Team?protocol6 min
- Agentic Coding Workflow Demands Context, Not Better Promptsengineering workflow6 min
- Apple Core AI and the Case for On-Device Inference at Workon-device ai5 min
- Write Better On-Call Handoff Notes With AI Doing the Draftingai-at-work5 min
- How LLM Tool Calling Works at the Token Levelai stack deep-dive6 min
- Why Does Your GitHub PR Review Still Live in Two Places?integration6 min
- Fine-Tuning vs Prompting for Narrow Tasks: When to Switchmodel fine-tuning6 min
- The AI Agent Deployment Gap Nobody Is Talking Aboutai agents6 min
- Understand the Nous Research Hermes Open-Weight Model Linefrontier models6 min
- The Slack Message That Sent an API Call Nobody Typedai fundamentals7 min
- Can Your AI Coding Agent Be Hijacked by a Bug Report?ai security5 min
- New Hire Slack Channel Onboarding, Done Without the Manual Workonboarding playbook6 min
- Hermes Desktop and the Self-Improving Open-Source Agentfrontier tools6 min
- Stop Picking AI Models From the Top of a Leaderboardmodel selection6 min
- Claude Tag puts an always-on AI teammate in Slacknews5 min
- Open Source AI Coding Agent OpenCode Hits 172k Stars as Copilot Bills by Tokencoding agents6 min
- Does Your AI Meeting Notes Tool Actually Close the Loop?ai-at-work5 min
- How AI Agent Memory Actually Works Under the Hoodai stack deep-dive5 min
- Figma Design-to-Developer Handoff, Honestly Assessedintegration6 min
- Pick an Open-Weight Model Based on What the Gap Data Showsfrontier models6 min