Web agents fail at hard real-world tasks

“Show HN: Agent-desktop – Native desktop automation CLI for AI agents I've been building computer-use tools for a while, and I quietly launched this about a month ago (122 Stars on GH). I figured it was worth sharing here. Over the last few months, a lot of computer-use agents have come out: Codex, Claude Code, CUA, and others. Most of them seem to work roughly like this: 1. Take a screenshot 2. Have the model predict pixel coordinates 3. Click x,y 4. Take another screenshot 5. Repeat That w”

“Show HN: Git for AI Agents hi guys. been working on something i think is fundamentally missing in today's workflow with ai agents. vcs. i find myself struggling with questions that agents can't answer like why did you do it? , when did u delete this folder? why? , etc. or trying to /rewind (after a /compact...) or basically `bisect` to find when and why something was done by the agent in the current / previous session. just like git did for code, i think we are the same ”

“Show HN: Kontext CLI – Credential broker for AI coding agents in Go We built the Kontext CLI because AI coding agents need access to GitHub, Stripe, databases, and dozens of other services — and right now most teams handle this by copy-pasting long-lived API keys into .env files, or the actual chat interface, whilst hoping for the best. The problem isn't just secret sprawl. It's that there's no lineage of access. You don't know which developer launched which agent, what it ac”

“Show HN: TinyFish Web Agent (82% on hard tasks vs. Operator's 43%) Enterprises need ~90% accuracy to deploy web agents. Until now, no agent has come close on real-world tasks. TinyFish is the first production-ready web agent. Here's the evidence. Results of hard task scores on Online-Mind2Web (300 tasks, 136 live websites, human-correlated judge): - TinyFish: 81.9% - OpenAI Operator: 43.2% - Claude Computer Use: 32.4% - Browser Use: 8.1% Why not WebVoyager like everyone else? Because it&#x2”

“Show HN: Broccoli, one shot coding agent on the cloud Hi HN — we built Broccoli, an open-source harness for taking coding tasks from Linear, running them in isolated cloud sandboxes, and opening PRs for a human to review. We’re a small team, and our main company supplies voice data. But we kept running into the same problem with coding agents. We’d have a feature request, a refactor, a bug, and some internal tooling work all happening at once, and managing that through local agent sessions meant”

“Show HN: rmBug – audited database access for humans and agents We've been building things together for a long time. LEGO first, then software. Across every company and project since, one thing kept showing up: database access security was broken. Not always dramatically. Sometimes it was the budget. Sometimes months of convincing. Sometimes just a quiet burden nobody talked about. Support staff with access to every customer's financial data. Engineers who left but somehow still had cre”

“Show HN: Mkdnsite – Markdown-native web server for humans (HTML) and agents (md) # What? Introducing mkdnsite ( markdown site ) - an open source Markdown-native web server that serves HTML to humans and raw Markdown to agents. No build step required. Runs on Bun/Node/Deno, as an OS-specific standalone executable, or as a Docker container. Possibly the easiest way to go from Markdown files to functional website in the new agentic era. Features: - Runtime-only, zero build - Content negot”

“Show HN: Agent-browser-shield – free extension to protect AI agents on the web I've been experimenting with Claude Code, ChatGPT Agent, and OpenClaw to perform more open-ended tasks for me online. A big blocker I've hit on shopping and research tasks is the agent getting a key piece of info wrong. For example, in one case, my agent decided to add a brand I don't like to the cart because the site flagged it as almost sold out The HN crowd is probably pretty aware of the threats and”

“Show HN: OpenSOP, We got tired of agents lying to us, so we built them a harness OpenSOP is an early open-source runtime/standard for executable agentic processes. You (or your agent) define a process in YAML, and OpenSOP exposes it as a typed REST API that agents and humans can both use. We built it because a lot of agent workflows still live in prompts, docs, or one-off scripts instead of versioned process definitions, and we wanted more control and auditability. Its under development, we”

“Show HN: MIT OSS LinkedIn DMs for Agents (CLI and Example TUI) I was tired of paying $100s/mo to access data I should own -- my own DMs on social media -- so I built Allman, a local-first cli to access linkedin messenger. Starting with LinkedIn, I gave the entire compiled js binary of linkedin's web app to claudecode and reversed engineered the entire messenger inbox in 24 hours. My goal is to bring this to all messengers so AI can handle all of this busywork, just like it can my email”