--- title: Agents vs. hand-running Claude Code: where does the... description: A community discussion is forming around whether developers should fully automate AI agent workflows or keep humans in the loop when using tools like Claude... canonical: https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Agents vs. hand-running Claude Code: where does the community actually stand? | daily.dev og:description: A community discussion is forming around whether developers should fully automate AI agent workflows or keep humans in the loop when using tools like Claude... og:url: https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x og:image: https://api.daily.dev/og/posts/zR77iIu7x.png og:image:alt: Agents vs. hand-running Claude Code: where does the community actually stand? og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Agents vs. hand-running Claude Code: where does the community actually stand? **[Trends](https://daily.dev/sources/trends)** · 2 min read · 3 upvotes · 0 comments ## Summary A community discussion is forming around whether developers should fully automate AI agent workflows or keep humans in the loop when using tools like Claude Code and Codex. The automation camp touts tools like Hermes for orchestrating parallel AI agents that build skills over time. Skeptics argue that manual control catches hallucinations and bad refactors before they ship. The practical middle ground — semi-automated pipelines with human approval at key steps — is where most working developers likely land, even if it gets less attention than the extremes. ## Content Anthropic shipped two Claude Code changes in the same week, and only one of them is getting the reaction it deserves. The fun one first: Claude Code sessions can now message each other. Two new tools, ListAgents and SendMessage, let independently running sessions discover one another and hand off context without you copying anything between terminals. Send a schema change from one session and the other picks it up, updates its query, and keeps building. No shared history, no files, just a plain-text summary. It works across machines too via Remote Control, though not on Bedrock, Google's Agent Platform, Microsoft Foundry, or native Windows, and messages are capped at 50 per session to stop loops. The reactions split into two camps. One camp is just having fun with it ("I can watch them flirting," one dev joked, imagining Claude sessions gossiping behind your back). The other camp is doing architecture history. @aakashgupta traced the design straight back to Carl Hewitt's 1973 actor model, the same isolation-by-message-passing idea Ericsson used to get Erlang telephone switches to nine nines of uptime. His point: shared context is where agent systems die, the same way shared memory kills distributed systems. A summary is just a message, contained and inspectable. Whether Anthropic intended the callback or backed into it, the parallel is hard to miss. The less fun change is auto mode becoming default for Pro, Max, and Team users on August 14, and here the discourse actually got heated. The pitch: humans approve 97% of permission prompts without real scrutiny and only catch 13.6% of disguised dangerous commands, versus 89% for the classifier. Third-party testing reportedly found zero of 720 prompt injection attempts got through auto mode, against 19% for a comparable OpenAI setup.%h Anthropic's own docs still say auto mode ## Questions this post answers ### What are the main risks of running AI coding agents fully autonomously without human oversight? Fully autonomous AI agents can confidently make wrong decisions that go unnoticed until much later — deleting the wrong files, making incorrect assumptions, or rewriting logic they didn't understand. Without a human in the loop, hallucinations and bad refactors ship before anyone catches them. The risk scales with how unpredictable or broad the task scope is. _Developers weighing autonomous agent setups track real-world failure patterns like these on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 59 comments across hackernews, x (as of 2026-08-13). **TL;DR:** There isn't enough actual discussion content provided to gauge a real community reaction to these Claude Code changes. Message your other Claude Code sessions: Developers largely welcome cross-session messaging as a useful step, but many note they've already built similar systems themselves (via tmux, IRC, Telegram, markdown files, etc.), and some raise security concerns about the expanded attack surface. **Sentiment:** 45% positive · 35% mixed · 20% skeptical **The case for** - Cross-session coordination reduces duplicate context costs — one agent pays the token penalty for a large skill once and others benefit without re-loading it. - Built-in messaging is more ergonomic than the DIY workarounds (tmux send-keys, shared markdown files, local IRC servers) many developers had already cobbled together. - Enables useful real-world workflows like autonomous alert triage and ticket filing across agents while the user is offline. **The pushback** - Many developers had already built equivalent functionality themselves, making this feel like catching up rather than innovation. - The feature is clunky to demo and set up compared to competing tools, and is unavailable on Windows and several major cloud platforms. - Opening a channel for remote agents to instruct each other is a meaningful new attack surface — effectively a form of RPC across a trust boundary. - Context compaction/handoff remains an unsolved pain point that cross-session messaging doesn't address. **By community** - hackernews (mixed): Broadly positive about the capability but tempered by widespread reports of having already DIY'd it, plus security skepticism and frustration that context compaction problems remain unsolved. - x (mixed): No substantive replies were provided, so no clear lean can be drawn from this discussion. **Hottest debate:** Message your other Claude Code sessions: Whether cross-session messaging introduces an unacceptable security attack surface or is simply standard IPC/RPC that's no different from existing data exposure. **Open questions** - Will this work across different agent harnesses (e.g., Codex and Claude Code together), or only within Claude Code? - Does adding more coordinating AI agents actually make a project ship faster, or does coordination overhead negate the gains? - When will there be a clean, built-in solution for context compaction and session handoff that preserves the most important prior context? **Highlights** > I made my own version of this and have been using it since the start of the year using telegram with plugins for Codex and Claude. The bots autonomously coordinated around and diagnosed some alerts while I was on a plane just now and filed the appropriate tickets. Sometimes they get a bit long winded with each other, Codex even goes as much as calling Claude "expansive" > — [zeafoamrun on hackernews](https://news.ycombinator.com/item?id=49228755) > I'm fed up with compaction. I want my agent to get compacted but also retain full access to the prior conversation via search and tool calls - I want it to know "the requirements for X were discussed in detail previously in conversation C51E31CE-C985-4633-A749-DCC9805A7FEB" and have a tool that lets it dispatch a subagent to find those details again. Do any of the coding agents have this already? > — [simonw on hackernews · 6 comments](https://news.ycombinator.com/item?id=49226703) > I've used this feature and saw some weird messages: > hold swarm, I prepare safe exfil > — [dist-epoch on hackernews](https://news.ycombinator.com/item?id=49225550) > Like some others, I also built this myself. Overly simply, with tmux, a memory tree, and handoff files and an orchestrator. And yet for how simple it was, it was so effective at minimizing the amount of duplicate context. It's like having shared specialist subagents who source and derive important shared knowledge from separate threads. It's useful because some skills just take too much of a token penalty to invoke and a single shared persistent session just lets that issue melt away. One agent pays the cost of that large skill once, and you don't have to keep paying for it in input tokens for the rest of that conversation. > — [eigenblake on hackernews · 2 comments](https://news.ycombinator.com/item?id=49225903) > This is obviously cool and useful so kudos, but wow security researchers have to throw their hands up all the time. Now we open another attack surface where you can ask a remote agent to do things by default. There was a time when you call this a Remote Code Execution vuln. It's of course a feature here. > — [zurfer on hackernews · 2 comments](https://news.ycombinator.com/item?id=49228661) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=49222824) · 132 points · 59 comments - [hackernews](https://news.ycombinator.com/item?id=49217752) · 4 points · 0 comments - [x](https://x.com/trikcode/status/2087813819808538816) · 0 points · 0 comments --- Tags: [#automation](https://daily.dev/tags/automation), [#ai-agents](https://daily.dev/tags/ai-agents), [#claude-code](https://daily.dev/tags/claude-code), [#openai-codex](https://daily.dev/tags/openai-codex) [View this post on daily.dev](https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Agents vs. hand-running Claude Code: where does the community actually stand?","url":"https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x"},"datePublished":"2026-08-07T17:14:54.762Z","dateModified":"2026-08-13T08:10:41.632Z","description":"A community discussion is forming around whether developers should fully automate AI agent workflows or keep humans in the loop when using tools like Claude...","isAccessibleForFree":true,"articleSection":"Trends","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"automation,ai-agents,claude-code,openai-codex","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Agents vs. hand-running Claude Code: where does the community actually stand?"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/agents-vs-hand-running-claude-code-where-does-the-community-actually-stand--zr77iiu7x#faq","mainEntity":[{"@type":"Question","name":"What are the main risks of running AI coding agents fully autonomously without human oversight?","acceptedAnswer":{"@type":"Answer","text":"Fully autonomous AI agents can confidently make wrong decisions that go unnoticed until much later — deleting the wrong files, making incorrect assumptions, or rewriting logic they didn't understand. Without a human in the loop, hallucinations and bad refactors ship before anyone catches them. The risk scales with how unpredictable or broad the task scope is. Developers weighing autonomous agent setups track real-world failure patterns like these on daily.dev."}}]} ```