> **Applied AI, Skills, and Claude Code · teamvince Writing** > Markdown representation of https://teamvince.com/writing > Short, opinionated essays on applied AI, Skills, agent harnesses, model routing, and what makes Claude Code actually stick inside real teams. Writing # Notes from the work. Short, opinionated pieces on applied AI, agent workflows, and the things that actually make internal tools stick. Curated ## Start here If you're new to the work, these three set the foundation. Foundation · 1 of 3 ### The Context Audit: why your team's AI tools aren't landing Your team has AI licenses and nobody's using them. That's not a training problem. It's a context problem. [Read the post →](https://teamvince.com/blog/the-context-audit) Foundation · 2 of 3 ### What I teach: a pattern from a dozen public builds A dozen public builds, one repeatable workflow underneath. The map from the builds to what the course teaches. [Read the post →](https://teamvince.com/blog/what-i-teach-through-public-building) Foundation · 3 of 3 ### How to write a lightweight CLAUDE.md (and AGENTS.md) Anthropic's own guidance is to keep CLAUDE.md short. Bloated files drag every conversation down. The 2026 lightweight pattern. [Read the post →](https://teamvince.com/blog/lightweight-claude-md-agents-md) Reading paths ## Pick the problem you're working on Each path moves from diagnosis to a concrete way to test the work. AI adoption · 3 posts ### Turn access into a workflow people use Start with the reason usage is uneven, choose evidence for one pilot, then practice on work the team can safely judge. 1. [Diagnose uneven adoption](https://teamvince.com/blog/ai-access-uneven-adoption) 2. [Measure the workflow](https://teamvince.com/blog/measure-ai-adoption-without-fake-roi) 3. [Run a safe working lab](https://teamvince.com/blog/safe-ai-labs-real-team-work) Reliable workflows · 4 posts ### Give agents context, boundaries, and proof Build the context layer first, spend human attention on consequential actions, then make the agent show a record you can inspect. 1. [Build durable context](https://teamvince.com/blog/why-context-is-the-only-ai-skill-that-compounds) 2. [Set risk-based approval](https://teamvince.com/blog/approval-fatigue-fake-control) 3. [Require a track record](https://teamvince.com/blog/agent-track-record-not-guardrails) 4. [Test eval-gated autonomy](https://teamvince.com/blog/earned-autonomy-eval-gated-agent) Themes ## Pillars The recurring threads across the writing. Pillar 01 ### Shipping with agents Agentic coding isn't magic. It's scope, specs, review, and verification. [Get the harness →](https://teamvince.com/harness?source=writing-pillar-shipping) Pillar 02 ### Context as the compounding layer AI adoption fails when tools lack durable context. Context is the only AI skill that compounds. [Start a consulting conversation →](https://teamvince.com/consult?source=writing-pillar-context) Pillar 03 ### Human review, evals, and reliability Don't trust outputs you can't inspect. Turn pass/fail checks, review gates, and lightweight evals into habits you can reuse. [See coaching →](https://teamvince.com/coaching?source=writing-pillar-reliability) Pillar 04 ### Build logs and case studies The site is a record of active shipping, not a static portfolio. [See the case studies →](https://teamvince.com/#projects) Pillar 05 ### Operator-led AI transformation The people closest to the workflow can now build the first version themselves. [Join the next cohort →](https://teamvince.com/waitlist?source=writing-pillar-operator) All posts - Sep 07, 2026 ### Loop Lab: The Exercise Behind the Agent Workshop Inside the Vonage and TechWalk agent workshop: the eight-part Loop Lab exercise, a downloadable contract, independent checks, and a clear first-run finish line. 6 min read - Aug 06, 2026 ### Don’t Loop the Latency in Voice AI Voice agents need a fast live path and slower offline loops. Learn where to draw the latency line, verify calls, and improve future behavior safely with Vonage. 6 min read - Jul 09, 2026 ### Your Team Has AI Access. Why Isn't Anyone Using It? AI licenses don't create adoption. Diagnose uneven team usage by checking workflow fit, context, manager signals, and the quality of the first useful result. 6 min read - Jul 08, 2026 ### How to Measure AI Adoption Without Inventing ROI Measure AI adoption with evidence your team can defend: baseline time, output quality, review effort, safety checks, and one clear decision after the pilot. 7 min read - Jul 07, 2026 ### How to Run Safe AI Labs on Real Team Work Run AI labs on real business workflows without creating a data mess. Use sanitized inputs, explicit review points, role-based tables, and a clear team handoff. 7 min read - Jul 06, 2026 ### Your Agent Doesn't Need More Guardrails. It Needs a Track Record. Teams deploying AI agents keep bolting on approvals and guardrails. The missing piece isn't oversight, it's a measured track record of what the agent earned. 6 min read - Jun 11, 2026 ### Earned Autonomy: An AI Agent That Earns Its Permissions Earned Autonomy is an AI support agent whose permissions are earned through evals, not configured: tiered trust, LLM-as-a-judge scoring, and automatic demotion. 8 min read - May 28, 2026 ### Every Approval Prompt Was Making You a Worse Operator Every approval prompt in Claude Code trains you to stop reading them. That's not safety, it's fake control. How risk-based approval fixes the wrong default. 5 min read - May 22, 2026 ### The Triple Debt Model: What Shipping Fast With AI Costs Your team ships faster than it can understand what it ships. The Triple Debt Model names the cost: technical, cognitive, and intent debt, in both directions. 5 min read - May 19, 2026 ### Three Things Anthropic Engineers Do With Claude Code Three habits from Anthropic engineers who use Claude Code daily: plans as HTML you can review, throwaway single-use tools, and a design system the model reads. 5 min read - May 14, 2026 ### What's an AI Agent? Explained One Slice at a Time What an AI agent actually is: a model with a goal, tools, and the freedom to loop until the job's done. Plus when you need a multi-agent system, by the slice. 6 min read - May 07, 2026 ### RefereeOS: An Evidence Board for Peer Review RefereeOS uses six AI agents and a Daytona sandbox to help peer reviewers verify preprints before bad science gets amplified. Built in 5 hours. First place. 7 min read - May 06, 2026 ### How to Write a Lightweight CLAUDE.md (and AGENTS.md) Anthropic's own guidance is to keep CLAUDE.md short. Bloated files drag every conversation down. The 2026 lightweight pattern for CLAUDE.md and AGENTS.md. 8 min read - May 04, 2026 ### Two Hours, Five Agents, No Orchestrator Built Commons Copilot at the Betaworks multi-agent hackathon in two hours. Five agents share one Intent Space. No queue, no planner, no dispatcher. 2nd place. 6 min read - Apr 21, 2026 ### Argue With Claude Before You Write Anything Built a sparring partner skill that pressure-tests my thinking before any writing starts. Most people use AI to write faster. Better to think harder first. 5 min read - Apr 14, 2026 ### I Taught Claude to Do My Monday Morning LinkedIn Homework Annoying weekly LinkedIn homework: export analytics, triage messages. So I taught Claude in Chrome how to do it. No scripts, no API, just patient teaching. 4 min read - Mar 31, 2026 ### The Context Audit: Why Your Team's AI Tools Aren't Landing Your team has AI licenses and nobody's using them. That's not a training problem. It's a context problem. A 30-day Context Audit ships the fix in four weeks. 6 min read - Mar 24, 2026 ### Why Context Is the Only AI Skill That Compounds Most AI tools fail inside teams because the tool doesn't know who's using it or what good looks like. Context is the only AI skill that compounds. 5 min read - Mar 17, 2026 ### Running Four Coding Agents in Parallel Ran four coding agents in parallel, each owning a case study end-to-end. The skill morphed from prompting to scoping the right work for the right tool. 5 min read - Mar 10, 2026 ### My Interview Stories Ran Long, So I Built Stellar Late-round interview, ADHD brain on the scenic route, interviewer flagged my stories as too long. Built Stellar that week: paste messy, get a clean STAR. 5 min read - Mar 03, 2026 ### What I Teach: A Pattern From a Dozen Public Builds A dozen public builds since November. One repeatable workflow underneath. That's what the course teaches. This is the map from the builds to the curriculum. 7 min read - Feb 24, 2026 ### A Multimodal Route Planner, Built in 24 Hours WalkRide combines walking, biking, and transit into smarter NYC routes. The 24-hour build log: why parallel APIs matter and what shipping fast taught me. 4 min read - Feb 17, 2026 ### I Built a Voice Agent for My Portfolio. V1 Was Embarrassing. An ElevenLabs recruiter visited my LinkedIn profile. Couldn't DM her back, so I built a voice agent for the next one. V1 was embarrassing. V2 wasn't. 7 min read - Feb 10, 2026 ### 11 AI Prompts That Run a Complete Job Search Got laid off. Spent 45 minutes per job application doing work a prompt could do in 5. Built an 11-prompt pipeline in two weeks, with a tracker to close it. 5 min read - Feb 03, 2026 ### From Restaurants to AI: What Hospitality Taught Me My path to AI building went through fine-dining floors in NYC and four and a half years in B2B SaaS. Most of what I know about shipping I learned on a floor. 7 min read - Jan 27, 2026 ### An ADHD Task System Powered by Multi-Agent AI ADHD-OS was my first serious multi-agent build, shipped as a Google/Kaggle capstone. Three agents, a constraint-aware design, and an April retrospective update. 6 min read Field notes ### Get the field notes Short updates when the agent workflow changes: what shipped, what broke, what I changed in the harness. > This page has a form a person fills in directly. Open the page to submit it. **You're on the list.** I'll send field notes when something useful changes in the workflow. No spam. I send when something useful changes. For teams ### Working on something specific? If a piece here mapped to something you're actually stuck on, I'd rather hear about it than guess at it from the outside. [Start a conversation](https://teamvince.com/consult) --- Markdown version of [https://teamvince.com/writing](https://teamvince.com/writing). Site context: [/llms.txt](https://teamvince.com/llms.txt) · [/agent-instructions.md](https://teamvince.com/agent-instructions.md) · [/sitemap.xml](https://teamvince.com/sitemap.xml).