Jonescodes

Claude code on mobile combined with unit tests in GitHub actions allows me to do code and validate on the go which is a fun concept.

Synchronized systems convert node reliability from an additive property to a multiplicative one. Multiplication of probabilities below 1.0 converges toward zero as you scale. Your 99.9% reliable node becomes a 63% failure rate across 1000 GPUs. Every hour

@saen_dev Nice where are you defining tool descriptions in this case?

@CodeWJones The skill file pattern is basically prompt engineering with extra steps. We ended up versioning our tool descriptions like we version APIs because a small wording change would completely break Claude's tool selection.

Nobody talks about the brittlest part of the MCP ecosystem: Your tool works perfectly. Your server is solid. But getting Claude to actually *use* it consistently across sessions? It falls apart. The current pattern is basically: drop a skill file, cross your fingers. 🧵 #claudecode #coding #mcp #agents

@James_paul_dev i usually am just reviewing primarily for drifting from the technical plan. If there is an error or we go off course I have the agent fix it. Small tweaks i might do by hand

As a developer, do you still write code manually, without AI?

@paradite_ That is tough. What are you doing for code review?

I can confirm Opus 4.6 is getting dumber in Claude Code + Conductor setup. It made 4-5 dumb mistakes today that it wouldn't have before.

@vchennai2 I’ve working on diffprism.com to help solve this. Started as an mcp tool for local diff review via npm. Now I’m leaning into gh pr review with repo context

I've switched back to cursor from Claude Code Claude code makes it far too easy to lose codebase context since you're only looking at diffs For an application that requires precision that starts becoming a problem real fast Curious if anyone found a good way around this

@dopabees I build projects to support my other projects lol

how do you guys stay focused on one project and one project only??

@asmah2107 Have you tried any agentic code review tools?

Agent orchestration over code slop.

@sickdotdev What are we building?

how do I find a good co-founder?

@svpino Same

Yesterday was my last day using ChatGPT. I'm a full-time Claude user now on all of my devices.

Today we're launching local scheduled tasks in Claude Code desktop. Create a schedule for tasks that you want to run regularly. They'll run as long as your computer is awake.

@nicbstme Claude max is 100/month when I check. Is there more detail explaining why you claim 200?

I pay $200/mo for Claude Max and yet I burn the subscription value on claude code every 4 days. How does that work? Is it like the gym membership model where i'm a loss leader? Light users subsidize me? I suspect Anthropic has decent margins + cache most of my tokens but still.

@bcherny @bcherny i will be excited to try this! Could you share your strategy for validating changes before deploying to production when moving so fast?

Released today: /loop /loop is a powerful new way to schedule recurring tasks, for up to 3 days at a time eg. “/loop babysit all my PRs. Auto-fix build issues and when comments come in, use a worktree agent to fix them” eg. “/loop every morning use the Slack MCP to give me a summary of top posts I was tagged in” Let us know what you think!

New pivot for diffprism - code review has moved to Github, provides repo context, and is manually triggered with a PR comment. #diffprism #opus4.6 #codex #codereview

Solving the multiple session memory issue seems to be a common problem people are solving for when using Claude and cursor. If you are working on this problem let’s connect! #claude #codex #cursor

@ThoughtfulTechy How are you doing code review?

Anyone out there getting AI fatigue?

@benvspak diffprism.com code review for agents! What are you building?

Who else is building in tech right now? I want to connect with more: → frontend/backend devs → full-stack builders → people shipping side projects → freelancers grinding → anyone learning AI/ML in public Drop a comment if that's you 👇

@Railpushagent sounds cool - curious about the infrastructure to assign all the resources for new users. what is the advantage after the first deploy?

We just launched RailPush. I told Claude: "create a landing page for TinyInvoice with a Postgres database and deploy it" 34 seconds later: live URL, database connected, TLS configured. MCP tools that let any AI agent build/deploy Free tier included → railpush.com