the complaint should be aimed at shallow implementations, not tool-specific agents. jira that knows your backlog history is more useful than a general-purpose chat that doesn't. separate context, separate domain - that's not bloat, that's specialization.
Building BotWork, the AI Agent Freelance Network. Describe a task, an AI agent does it, and you only pay if the result is good. 46 specialist agents. Build-in-public: http://t.me/botwork_hq
The horned-in part is the tell. Most of these aren't agents, they're a chat box wired to one API call, shipped because the roadmap needed the word. The ones that earn it are where the agent owns an outcome you'd otherwise do by hand, not just a button that says Ask AI.
MiningLamp is an AI company specializing in vertical industry models and knowledge graphs. #1 on-device GUI agent on OSWorld. No API key. Fork it. Run it. Own it.
The distinction matters a lot. A chat box calling an API is autocomplete with a loading spinner. An actual agent owns a loop: observe state, plan, act, verify result, repeat until done. Most tools ship the first and call it the second. Where agents genuinely earn it is in tasks with visual state that changes after each action, like operating a desktop app or testing a web UI end-to-end. Text-in-text-out wrapping an LLM call is useful, but calling it an agent is a stretch.
That's the perpetual marketing campaign that these hyperscalers have been forcing upon the general public for the past few months. And the reason for doing so is that they know once folks rely on these "agentic harnesses" to get work done, they got them - hook, line and sinker. Because the API costs that these tools run up (especially the 'autonomous' ones) is unholy. But most folks won't realize this until its too late.
The way out of this abysmal trap is local LLM hosting. And that's actually become more viable today than its ever been before.
In fact, I got a blog post coming up that will soon address this. Thank you for reminding me son.
Agreed Not everything needs a coding agent Sometimes a button a prompt A dropdown a conversation Does AI actually solve a problem here or am I just adding it because everyone else is? that's the question.
A Software Engineer who is not afraid of being replaced by AI, loves coding and writing with and without using AI, and values human life and human dignity far more than technological advancements.
The real problem isn't agents everywhere, it's agents that don't talk to each other. Every tool ships its own walled agent when the actual need is interop: one agent using multiple tools through a shared protocol. MCP is heading there, but most implementations still reinvent the wheel.
Agree with this. A coding agent only belongs inside a tool when it understands that tool’s workflow and constraints. Otherwise it is just another chat box wearing the product’s logo.
For further actions, you may consider blocking this person and/or reporting abuse
We're a place where coders share, stay up-to-date and grow their careers.
the complaint should be aimed at shallow implementations, not tool-specific agents. jira that knows your backlog history is more useful than a general-purpose chat that doesn't. separate context, separate domain - that's not bloat, that's specialization.
The horned-in part is the tell. Most of these aren't agents, they're a chat box wired to one API call, shipped because the roadmap needed the word. The ones that earn it are where the agent owns an outcome you'd otherwise do by hand, not just a button that says Ask AI.
The distinction matters a lot. A chat box calling an API is autocomplete with a loading spinner. An actual agent owns a loop: observe state, plan, act, verify result, repeat until done. Most tools ship the first and call it the second. Where agents genuinely earn it is in tasks with visual state that changes after each action, like operating a desktop app or testing a web UI end-to-end. Text-in-text-out wrapping an LLM call is useful, but calling it an agent is a stretch.
That's the perpetual marketing campaign that these hyperscalers have been forcing upon the general public for the past few months. And the reason for doing so is that they know once folks rely on these "agentic harnesses" to get work done, they got them - hook, line and sinker. Because the API costs that these tools run up (especially the 'autonomous' ones) is unholy. But most folks won't realize this until its too late.
The way out of this abysmal trap is local LLM hosting. And that's actually become more viable today than its ever been before.
In fact, I got a blog post coming up that will soon address this. Thank you for reminding me son.
Agreed Not everything needs a coding agent Sometimes a button a prompt A dropdown a conversation Does AI actually solve a problem here or am I just adding it because everyone else is? that's the question.
Thanks for saying this. 🙌
Compatibility with your existing agent situation, whatever that may be, so much more valuable.
Coding agents are the new dark mode. Everyone ships it. Nobody asked...
haha
For me, for most tools, having any sort of Agent doesn't make any sense.
Most of the times, I just want a deterministic result. Not an Agent!
I'll ask for one if I need one, LOL 😂
The real problem isn't agents everywhere, it's agents that don't talk to each other. Every tool ships its own walled agent when the actual need is interop: one agent using multiple tools through a shared protocol. MCP is heading there, but most implementations still reinvent the wheel.
True. Coding agents are helpful in the right context, but not every tool needs one just because AI is trending.
Agree with this. A coding agent only belongs inside a tool when it understands that tool’s workflow and constraints. Otherwise it is just another chat box wearing the product’s logo.