Today's win: after 6 months of building solo, Convosparr is live in production. The problem it solves: salespeople and founders get no safe reps. You learn to handle objections by fumbling them in front of real prospects. Every practice call costs you real pipeline. What's live today: Live voice practice calls. Not text, not multiple choice. You speak, an AI buyer speaks back in real time, and it feels uncomfortably close to a real prospect. It interrupts, it pushes back, it loses interest if you waffle. Real scenarios. Cold calls, pitch meetings, objection handling. Each with difficulty levels, so the buyer can be warm and curious or busy and borderline hostile. Your own context. Brief it on your product and who you're calling, and you're rehearsing tomorrow's actual meeting instead of a generic script. Scored analysis after every call. A breakdown of how the conversation actually went: opening, discovery, objection handling, closing, with specific things to fix. Then you call again and try to beat it. I built this alone, no funding, no team, because I believe practicing on real prospects is the most expensive way to learn sales. It's at convosparr.com and there's a free tier. Try one call. Then come back and tell me what a real buyer would have done differently. That feedback is exactly what I need right now.

Hey everyone, I've been building a voice AI sales training platform and I'm looking for a handful of beta testers to help me figure out if this is worth pursuing further. What it does: You hop on a live voice call and practice cold calls and pitches against AI personas that respond like real prospects, pushing back, asking hard questions, and keeping you on your toes. After each session you get a scored breakdown of your performance. That's the core of it right now. It's early. Cold calls and pitches are what's live, and I want real feedback from people who actually do this work before building anything else. Who I'm looking for: Anyone with sales experience. SDRs, AEs, founders doing outbound, or anyone actively making calls and pitches. The more real-world reps you've done, the more useful your feedback will be. What you get: - Free access during the beta - Direct input on whether this is something worth building out further If you want in, comment below and I'll reach out to get you set up. No credit card, no fuss. Thanks! Karthik

Hello Everyone, My name is Josh and I have been working in tech for over a decade. I currently work for one of the largest cloud companies in the world helping my clients embrace AI and cloud. Looking forward to learning and connecting from this community.

Just found one of the coolest “vibe coded” projects I’ve seen in a while: 🌆 A 3D city where every GitHub developer is a building How it works: - More commits = taller building - More repos = wider base - Lit windows = recent activity It’s such a smart way to turn boring profile stats into something visual + instantly understandable. 🔗 Check it out: thegitcity.com This is the kind of project that reminds you: You don’t need to build something “serious” to get attention. You need to build something people instantly get and want to share.

I’m seeing a recurring pattern with long-running agents: a lot of failures aren’t because the model is “bad”, but because the workflow is packaged poorly. If an agent is held together by one giant prompt, it tends to drift, forget context, and turn into “prompt spaghetti.” The cleaner approach (from OpenAI’s Skills + Shell + Compaction framing) is to treat the workflow like software: procedures live in a skill, execution happens in a real environment (Shell), and context gets compacted so runs can continue without falling apart. A detail that stood out: Glean.ai saw skill routing initially drop by about 20% in evals, largely because descriptions weren’t written like routing logic. The takeaway is simple: skill descriptions should be decision boundaries, not marketing copy. A few practical habits I’m adopting: write clear “use when / don’t use when” rules, add negative examples when skills can be confused, keep templates inside the skill (not the system prompt), and when reliability matters, explicitly instruct “Use the <skill name> skill.” There’s also a solid datapoint in the post: a Salesforce-oriented skill example improved eval accuracy from 73% to 85%, and time-to-first-token dropped by 18.1%. Curious: what’s your biggest failure mode with long-running agents right now?