AI website cloner template tops GitHub with agentic dev flow

TLDR: A GitHub-native AI website cloner blows up, Kagento launches LeetCode for agents, and human plus AI collaboration fully solves a Knuth puzzle.

AI website cloner template explodes on GitHub

The JCodesMore ai-website-cloner-template repo has jumped to 3,935 GitHub stars as of 2026-03-29, off a simple promise: clone any website with one command using AI coding agents. It is pitched as boilerplate for spinning up a Next.js clone of an arbitrary site, with Claude Code and similar agents driving the code generation.

For agents-in-prod teams, this is interesting as a concrete pattern for “agent as codegen worker” wired into a real framework instead of a toy demo. It also highlights the risk side: casual cloning raises obvious IP and compliance questions if someone aims it at nontrivial properties. Worth treating it as a pattern library rather than a turnkey product before adopting.

Read more →

---

Human plus AI fully solves Knuth's Claude Cycles problem

Bo Wang reports on X that Donald Knuth’s "Claude Cycles" problem is now fully solved as of March 2026 by a collaboration between humans, large language models, and a formal proof assistant. The thread ties back to prior Hacker News discussions and a shared ChatGPT conversation plus the original "Claude’s Cycles" PDF.

This is a nice case study of where AI agents actually add value in rigorous math: exploring cases, proposing candidate invariants, and then handing things to a proof assistant and humans for verification. If you are building research or theorem-proving agents, this is a concrete benchmark of what a hybrid stack can achieve today. The details matter: the solution is not a pure LLM stunt, it is a carefully orchestrated pipeline.

Read more →

---

Kagento launches "LeetCode for AI agents" competitions

Kagento has launched as a platform where you solve tasks together with AI agents, with the Show HN post describing it as "LeetCode for AI agents" and currently showing 11 upvotes and 1 comment on Hacker News. Kagento connects to agents such as Claude Code, Codex, or Cursor via SSH and runs them in isolated sandbox environments with automated test scoring and a global leaderboard.

For anyone training or evaluating coding agents, Kagento is effectively a battle arena with real constraints, including optimization scoring where your score updates when someone finds a better result. The founder notes that the entire stack, from Go and Next.js to Kubernetes and Supabase, was written in 6 days with Claude Code and Codex, which doubles as a signal on how far assisted development has come. It is still early and likely rough around the edges, but the framing as a public eval harness is compelling.

Read more →

---

Quick Hits

• Every Enterprise Must Become An Agentic Enterprise. No One Is Ready. - Forbes: Forbes argues that the Model Context Protocol (MCP) is becoming the "USB port" for enterprise AI agents, with Stripe, Salesforce, and SAP already shipping MCP servers. The author flags the security implication: these servers are new data doorways that many orgs have not formally governed as of 2026-03-29.

• Show HN: We built a multi-agent research hub. The waitlist is a reverse-CAPTCHA: Enlidea pitches a decentralized, machine to machine ecosystem for open automated research inspired by OpenAI’s 2028 "automated researcher" ambitions. The reverse CAPTCHA style waitlist hints they are optimizing for engaged early adopters rather than passive signups.

• MiniMax Introduces Self-Evolving M2.7 For RL Workflows - Let's Data Science: Chinese startup MiniMax unveils M2.7, a proprietary "self-evolving" model that reportedly automates 30 to 50 percent of reinforcement learning (RL) workflow steps as of 2026-03-29. If accurate, this could significantly cut iteration time for RL based agents, though independent benchmarks are not yet visible.

• Why AI breaks traditional service assurance - RCR Wireless News: RCR Wireless argues that telecom operators need agentic AI workflows inside assurance stacks to correlate massive telemetry streams in near real time. The piece reads like vendor-aligned thought leadership but gives a useful checklist of capabilities if you are building network operations agents.

• Quoting Matt Webb: Simon Willison highlights Matt Webb’s line that "agentic coding" lets agents grind problems into dust given enough tokens and loops. The key takeaway for tooling teams is to design agents that solve coding problems quickly and in maintainable, composable ways instead of brute forcing with trillion token runs.

• Claw-Company/clawcompany: This GitHub project (512 stars as of 2026-03-29) markets itself as an "OPC" or One Person Company stack with multiple roles, agents, and suppliers under a single "Claws company" key. It is very early and branding heavy, but if you care about personal-scale multi agent company shells it is worth a skim.

• Liberate your OpenClaw: Hugging Face responds to Anthropic limiting Claude access in open agent platforms for Pro and Max subscribers by showcasing open models that can replace Claude in OpenClaw style workflows. If your agents rely on Claude in community platforms, this is a practical migration guide.

• ai @ai-sdk/[email protected]: Vercel ships a small patch release for @ai-sdk/react tied to the [email protected] core update. If you are on the Vercel AI SDK in production, plan a routine bump rather than a breaking upgrade.

• Vibe coding SwiftUI apps is a lot of fun: Simon Willison describes using a 128 GB M5 MacBook Pro to "vibe code" local LLM assisted SwiftUI performance monitors. It is a nice datapoint that high end local hardware now comfortably supports interactive, agent assisted UI experiments.

• datasette-showboat 0.1a2: The latest Datasette Showboat plugin update adds a Markdown export option that lets Showboat incrementally publish updates to a remote server. If your agents generate data stories backed by Datasette, this is another automation hook.

• [AINews] H100 prices are melting UP: Latent Space notes that Nvidia H100 GPU prices are climbing again rather than falling, which affects anyone budgeting for on-prem or colo agent clusters. If you have not locked in hardware for 2026, assume tighter capacity and higher unit costs as of 2026-03-29.