# Coding Agents: zed.dev Source: https://paragraph.com/@papajams.eth/zed-dot-dev Kiwi News: https://news.kiwistand.com/stories/Coding-Agents-zeddev?index=0x685b0660c56ca1057d86fc2bcd4bc946911a4035fb500296a2a880973b6011b4d9253cf7 Submitted by misha on 2025-06-24 UTC, 17 upvotes, 4 comments ## Summary The author, who writes under the handle papa, argues that AI-integrated tooling, using Zed as the example, lets one developer build full-stack systems in weeks that previously needed a team of five over six months. Two projects are described in detail: MegaVibe, a decentralized tipping platform with Mantle blockchain micropayments and GPS-verified proof-of-presence, which the author says was delivered as an MVP in 1 week for under $500 in infra and tooling costs; and ImperfectCoach, an AI coaching app built on OpenAI, Gemini, and Claude, created in under 4 weeks. The article also includes a comparison table claiming Zed starts in under one second and uses about 150MB of RAM, versus 3-5 seconds and about 500MB for VS Code. ## Article I Built Three Apps Solo in Weeks. Two years ago, I had this idea amongst many others, but was stuck in a familiar loop: overly ambitious prototypes leading to abandoned repos. Integrating machine learning? Needed an expert. Blockchain micropayments? Complexity rabbit hole. Real-time computer vision and audio processing? WTF. Fast forward to 2025 and—i'm shipping. No team. No burnouts. No funding. The shift? AI-integrated tooling. I no longer have to choose between ambition and scope. With tools like Zed, I can build full-stack systems with the potential to bring my vision to reality. This isn’t a productivity hack. It’s a structural shift in how software is made. Traditional development follows a linear growth model: more features = more time = more people. But exponential developers operate differently. With AI-native development environments, they: Offload complexity to AI agents Rapidly prototype across technical domains Maintain architectural integrity without mental overhead What used to take a team of five over six months can now be done by one person in a few weeks. The rise of the exponential developer sits at the intersection of: Wright’s Law: Efficiency improves with cumulative experience Moore’s Law: Computing power doubles roughly every two years Kurzweil’s Law of Accelerating Returns: Technological change is exponential, not linear We’ve hit an inflection: where AI doesn’t just assist code completion—it drives full system architecture. Zed's UI/UX modifications span animations, feature cards & much more Idea: A platform where artists monetize performance energy through real-time tips, verified by location. Architecture: Mantle blockchain micropayments GPS-verified proof-of-presence protocol WebRTC live streaming Decentralized tipping infra Old Way: 4-5 developers across frontend, blockchain, backend, and DevOps 8-12 months development $500K+ budget What I Did Instead: With Zed, I: Deployed smart contracts in days, guided by AI-generated Solidity patterns Built GPS proof-of-presence logic with AI code review support Integrated WebRTC livestreaming using AI-guided React flows Delivered the MVP in 1 week: https://megavibe.vercel.app Results: Deployed a fully working decentralized tipping platform Shipped live to users for under $500 in infra/tooling costs (testing in beta at next event) Goal: Real-time AI coaching with LLM orchestration and user motion tracking Stack: OpenAI, Gemini, Claude via serverless LLM router Real-time motion tracking with MediaPipe Blockchain-based identity verification Supabase DB + AWS Lamda Traditional Build Requirements: AI/ML expert, blockchain dev, backend architect, UI/UX engineer 6-9 month timeline My Reality: Zed’s multibuffer editing enabled orchestration of multiple LLMs AI agent helped debug API response issues across providers Created the full system from scratch in under 4 weeks: https://imperfectcoach.netlify.app Key Unlocks: Zero external research required for AI model integration Real-time code insights into system dependencies Single codebase orchestration with full architectural visibility AWS + Supabase Backend Support There is ZERO chance I would ever have been able to navigate the complexity of AWS console + Supabase simultanously to deliver on this. Its performant and useful, super proud its out. For a limited time its free for users to try out, kindly share feedback and tweet about us: https://x.com/ifdotfun Vision: Personalized breath coaching using computer vision and real-time audio processing Technical Stack: Computer vision for posture analysis Real-time audio analysis for breath pattern detection AI-generated coaching feedback WebRTC for live coaching Challenges: Real-time ML pipelines Cross-modal data synchronization Frontend motion design My Approach: Built CV pipeline in 48 hours with AI assistance Designed ML-based feedback loop without ML expertise Streamlined complex UI animation using AI-injected CSS/JS logic Output: A functional, responsive platform demo in a week Ready for real user feedback and live testing ZED ui/ux is beautiful, as is its efficacy! Factor Traditional Team Exponential Dev (Me) Timeline 8-12 months 3-4 weeks Team Size 4-6 people 1 Cost $500K+ <$20K Tooling VS Code + plugins Zed (AI-native) Debugging Time 40+ hrs/week ~10 hrs/week (AI-aided) These are not minor improvements. They're a restructuring of the cost-function of software development. When I first came across it I was hesitant to download a dedicated IDE - why not just use an extension in VS CODE?! Turns out by vertically integrating their stack they can really optimise the user experience. Key Differentiators: <1s startup time Native multiplayer coding Context-aware AI via Agent Panel Low memory + power usage Full-codebase understanding Table: Zed vs Others Metric Zed VS Code JetBrains Startup Time <1 second 3-5 sec 5-10 sec RAM Usage ~150MB ~500MB ~700MB Collaboration Built-in Plugins Complex AI Context Full Partial Plugin Source Model Open (GPLv3) Closed Closed AI as a Teammate: Not a helper. I ask it hard questions. We design systems together. Architectural Thinking Wins: Coding is easier now. The hard part is systems thinking. Start Small, Iterate Fast: I don't aim for perfect. No problem is now unsolvable. Imagination Is the New Bottleneck: Not tech skills. The hardest part of building used to be execution. Now, it’s ideation and discernment. If you’re a developer who’s been burned by the complexity of full-stack ambitions, I’m here to tell you: it’s time to build again. The tools have changed. The game has changed. And you? You just might just be a more exponential developer than I am, if you try! All projects referenced in this article are live open-source repositories available on GitHub: MegaVibe ImperfectCoach ImperfectBreath Supporting writers is beyond the call of duty, reading this far is its own form of glory, appreciate you. Farcaster @papa — warpcast.com/@papa Lens @papajams — lenster.xyz/u/papajams Twitter @papajimjams — twitter.com/papajimjams PAPA - runwayML: https://paragraph.com/@papajams.eth/from-photos-to-claymation-with-runwayml PAPA - lens hackathon: https://fountain.ink/p/papajams/p0psmk45tcrp0c1rgd PAPA - filecoin hackathon: https://paragraph.com/@papajams.eth/filecoin-hackathon PAPA - mento hackathon: https://paragraph.com/@papajams.eth/global-stablecoin-hackathon ## Comments **tomw1808.eth**: Is that an ad for zed?! I do acknowledge the technical depth of the Prototypes, but all linked prototypes severely lack UI parts and really look prototypes, but not from 2025 but prototypes from 2005. Arguably yes, its a prototype, but really, broken menubars, broken animations, entirely missing loading states or start-states, broken UI design and overflowing containers left and right. What I am trying to get at is that its really hard to do: with any of the newer models its actually really hard to make it look like that. The models are insanely good at applying best practices, where it looks almost a bit too fantastic out of the box. Especially in well known languages such as JS/CSS. I assume it was either a really stringent prompt that led to this result, or maybe just general lack of framework usage? **misha**: Believe it’s not an ad ;) I found it through a post of @papajams on Hey - I think he’s the author also. He’s an active poster and pro user. But maybe ask him about it ;) https://hey.xyz/posts/74kb9bfk5bmp75jp8a **tomw1808.eth**: okay, now comes an embarrassing question: what is hey.xyz and why is everyone jumping on it suddenly? What am I missing here? **misha**: Just checked the Hey Dune metrics and indeed signups have skyrocketed recently. It has probably to do with the launch of Hey Pro. https://dune.com/heyverse/hey Anyway hey.xyz has been around for a while. It’s one of the instances of Lens lens.xyz , which has now their own L2 chain, Lenschain. Another instance for mobile is Orb https://orb.club , which has just been bought by Mask Network https://www.mask.io (which is even older than Lens). Anyway Lenschain aims to be the social chain :) Avara, the company behind Aave, Lens, and Family is where most of the investing comes from.