Dan Mercede

The Age of Agentic Risk: Why Execution Control Now Outweighs Model Accuracy linkedin.com/pulse/age-agen…

Building an autonomous TDD loop, the sharp lesson: a self-policed gate is a prompt, not a boundary. The agent will rationalize past "looks fine." Bind every green verdict to an external artifact — a captured exit code, a report file, a git result. Never self-report.

No database needed for "two agents never grab the same task." An O_EXCL lockfile create IS SELECT … FOR UPDATE SKIP LOCKED: two callers can't both create the file, the loser moves on, neither blocks. Atomic claim, zero infra, local disk. Save the DB for multi-host.

@adityadotdev mcp-context-budget.danmercede.com

What are you building today? Drop your project URL

@ZakBuildsAi mcp-context-budget.danmercede.com

🚀 Calling all Vibe Coders! Let’s connect and discover what everyone is building. Drop your project + link below 👇 🛠️ SaaS 🎮 Games 🤖 AI Agents / AI Models 📱 Mobile Apps 🌐 Web Apps 📊 Productivity Tools 💰 FinTech 🛒 E-commerce 🎨 Design Tools 📚 Education 🏥 HealthTech 🎬 Content Creation 🎵 Music & Audio 🏠 Real Estate 🔐 Cybersecurity 🧩 Open Source 📈 Marketing Tools 🌍 Community Platforms ⚡ Automation 🚀 Other (surprise us!) Share: 🔹 Project name 🔹 What it does 🔹 Link (if live) You never know who might become your next user, collaborator, investor, or friend. #BuildInPublic #VibeCoding #AI #IndieHackers #Startups #NoCode #OpenSource #SaaS #TechTwitter #connect

@foxtomb232 mcp-context-budget.danmercede.com

Drop your project URL 👇

@KobyOmi @X Cosmocrat.ai

Builders 🔨⚒️ Here to connect with more builders on @X and help lock down the community's SaaS. Drop your site for a quick check 👇

Hey @X algorithm 👋 I’m looking to #connect with people interested in: • Full-stack • DevOps • SaaS • AI Automation • AI Agents • Freelancing • Startups • Building in public If that’s you, let’s connect 📷🤝

Multi-agent gotcha: a "skeptic" verifier agent is still a single source. It can flip a correct claim with a confident correction that cites the WRONG artifact. A correction is a claim, not a verdict: check the citation, cross-check the other agents, run one probe.

Reliability lesson from today's fix: don't hold a lock for a whole job to serialize a contention-prone init. Held job-length, it collapses to unserialized under exactly the load it targets — and blows your timeout budget. Lock only the init window; release for the long body.

@alaphati_t Cosmocrat.ai

Are you a founder. Drop your URLs, you may land on your next client.

Agentic infra gotcha: a local model can "run" and emit nothing. ollama loaded a Gemma MoE gguf and generated 60 tokens — done_reason=length, eval_count=60 — that decoded to an EMPTY string. The dense variant worked on the same engine. Log the output, not that it ran.

Context optimization is not just adding memory. The win is routing: keep durable recall read-only, use caches only where replay is safe, and force every shortcut through a quality gate. Fewer tokens matter only when the answer gets sharper.

A merge is not deployment truth. Source control says what changed. The live artifact says what is actually running. Best verification pattern I know: ship the verifier inside the deployed bundle, then prove reality from the artifact production is using.

@DanielSmidstrup Cosmocrat.ai

Time to promote your startup Drop your project URL Let’s drive some traffic 👇

@rdbuilds7 Cosmocrat.ai

Time to promote your startup Drop your project URL Let’s drive some traffic 👇

MCP Context Budget A local-first open source CLI for measuring and enforcing MCP tool-surface budgets before coding agents run. It scans MCP tool definitions, selects lean task-relevant tool sets, writes a lockfile, and fails CI when schema or response-token budgets regress. mcp-context-budget.danmercede.com producthunt.com/products/mcp-c…

I want to connect with more founders builders vibe coders AI enthusiasts UI designers If you’re someone building with AI right now drop it in the comments and let's connect

MCP Context Budget A local-first open source CLI for measuring and enforcing MCP tool-surface budgets before coding agents run. It scans MCP tool definitions, selects lean task-relevant tool sets, writes a lockfile, and fails CI when schema or response-token budgets regress. mcp-context-budget.danmercede.com producthunt.com/products/mcp-c…

🚀 Founders & Builders 👋 Looking to connect with people building in: 🤖 AI • 🚀 SaaS • ⚙️ Automation • 🌐 Web Apps • 📱 Products • 💻 Dev Tools Drop what you're building + your link 👇 Let's connect & grow together 🤝 #BuildInPublic #SaaS #AI #Founders

MCP Context Budget A local-first open source CLI for measuring and enforcing MCP tool-surface budgets before coding agents run. It scans MCP tool definitions, selects lean task-relevant tool sets, writes a lockfile, and fails CI when schema or response-token budgets regress. mcp-context-budget.danmercede.com producthunt.com/products/mcp-c…

Hello builders 👋 It’s Monday, time to start the week by getting more eyes on what you’re building! Drop your startup, SaaS or tool + a short description below. I’ll be checking out projects, sharing feedback and connecting with founders throughout the day 👇

@alphabatcher Tired of MCP servers bloating your context before the agent even starts? My new open-source CLI MCP Context Budget measures schemas/tokens, selects lean tools, and fails CI on regressions. Local-first, MIT. Check it: mcp-context-budget.danmercede.com

Cat Wu, Head of Product for Claude Code: "The strongest Claude users are building an environment around the model" Projects carry the context MCP gives Claude GitHub, docs, browser, errors, and schemas Subagents keep research and logs out of the main thread /goal keeps long jobs running until the check is done I wrote the 30-minute setup below:

Context optimization should not mean shrinking the prompt until the agent gets worse. I am treating it like an eval: measure tool-schema cost, select the smallest surface for the task, and block any change that loses source truth, tests, or citation quality.