How to build a coding agent (ghuntley.com)

We (the Princeton SWE-bench team) built an agent in ~100 lines of code that does pretty well on SWE-bench, you might enjoy it too: https://github.com/SWE-agent/mini-swe-agent

Anyone can build a coding agent which works on a) fresh code base b) when you've unlimited token budget

now build it for old codebase, let's see how precisely it edits or removes features without breaking the whole codebase

lets see how many tokens it consumes per bug fix or feature addition.

[dead]

I really think the current trend of CLI coding agents isn't going to be the future. They're cool but they are _too simple_. Gemini CLI often makes incorrect edits and gets confused, at least on my codebase. Just like ChatGPT would do in a longer chat where the context gets lost: random, unnecessary and often harmful edits are made confidently. Extraneous parts of the codebase are modified when you didn't ask for it. They get stuck in loops for an hour trying to solve a problem, "solving it", and then you have to tell the LLM the problem isn't solved, the error message is the same, etc.

I think the future will be dashboards/HUDs (there was an article on HN about this a bit ago and I agree). You'll get preview windows, dynamic action buttons, a kanban board, status updates, and still the ability to edit code yourself, of course.

The single-file lineup of agentic actions with user input, in a terminal chat UI, just isn't gonna cut it for more complicated problems. You need faster error reporting from multiple sources, you need to be able to correct the LLM and break it out of error loops. You won't want to be at the terminal even though it feels comfortable because it's just the wrong HCI tool for more complicated tasks. Can you tell I really dislike using these overly-simple agents?

You'll get a much better result with a dashboard/HUD. The future of agents is that multiple of them will be working at once on the codebase and they'll be good enough that you'll want more of a status-update-confirm loop than an agentic code editing tool update.

Also required is better code editing. You want to avoid the LLM making changes in your code unrelated to the requested problem. Gemini CLI often does a 'grep' for keywords in your prompt to find the right file, but your prompt was casual and doesn't contain the right keywords so you end up with the agent making changes that aren't intended.

Obviously I am working in this space so that's where my opinions come from. I have a prototype HUD-style webapp builder agent that is online right now if you'd like to check it out:

https://codeplusequalsai.com/

It's not got everything I said above - it's a work-in-progress. Would love any feedback you have on my take on a more complicated, involved, and narrow-focus agentic workflow. It only builds flask webapps right now, strict limits on what it can do (no cron etc yet) but it does have a database you can use in your projects. I put a lot of work into the error flow as well, as that seems like the biggest issue with a lot of agentic code tools.

One last technical note: I blogged about using AST transformations when getting LLMs to modify code. I think that using diffs or rewriting the whole file isn't the right solution either. I think that having the LLM write code that modifies your code and then running that code to affect the modifications is the way forward. We'll see I guess. Blog post: https://codeplusequalsai.com/static/blog/prompting_llms_to_m...

Where is the program synthesis? My way of thinking is given primitives as tools, i want the model to construct and return the program to execute.

Of course following nix philosophy is another way.

Why are any of the tools beyond the bash tool required?

Surely listing files, searching a repo, editing a file can all be achieved with bash?

Or is this what's demonstrated by https://news.ycombinator.com/item?id=45001234?

Instead of writing about how to build an agent, show us one project that this agent has built.

For me, the post is missing an explanation of the reason why I would want to build my own coding agent instead of just using one of the publicly available ones.

A very similar "how to guide" can be found here https://ampcode.com/how-to-build-an-agent written by Thorsten Ball. In general Amp is quite interesting - obviously no hidden gem anymore ;-) but great to see more tooling around agentic coding being published. Also, because similar agentic-approaches will be part of (certain/many?) software suits in the future.

I hate to do meta-commentary (the content is a decent beginner level introduction to the topic!), but this is some of the worst AI-slop-infused presentation I've seen with a blog post in a while.

Why the unnecessary generated AI pictures in between?

Why put everything that could have been a bullet point into it's own individual picture (even if it's not AI generated)? It's very visually distracting, breaks the flow of reading, and it's less accessible as all the picture lack alt-text.

---

I see that it's based on a conference talk, so it's possibly just 1:1 the slides. If that's the case please put it up in it's native conference format, rather than this.

Building a coding agent involves defining clear goals, leveraging AI, and iterating based on feedback. Start with a simple task and scale up.

If a picture is usually worth 1000 words, the pictures in this are on a 99.6% discount. What the actual...?

what's the best current cli (with a non interactive option) that is on par with Claude code but can work with other llms like ollama, openrouter etc? I tried stuff like aider but it cannot discover files, the open source gemini one but it was terrible; what is a good one that maybe is the same as CC if you plug in Opus?

Exactly my approach to gaining knowledge and learning through building your own(`npx genaicode`). When I was presenting my work on a local meetup I got this exact question: "why u building this instead of just using Cursor". The answer is explained in this article(tl;dr; transformative experience), even though some parts of it are already outdated or will be outdated very soon as the technology is making progress every day.

The problem I have with this is that this style of agent design, providing enormous autonomy, makes sense in coding while keeping an expert human in the loop since it can self-correct via debugging. What would the other use cases of giving an agent this much autonomy be today versus a more structured flow versus something more like LangGraph?

Very simplistic view on the problem domain IMHO. Yah sure we can add a bunch of functions... ok. But how about snapshotting (or at least work with git), sandboxing both process and network level, prompt engineering, detect when stuck, model switching with parallel solvers for better solutions. These are the kind of things that make coding agents reliable - not function declarations.

The trick with coding agent is guiding the attention towards tasks it can expect will fit in the agent’s token window and deciding when to delegate. Funny as a PM you have the exact problem.

> You just keep throwing tokens at the loop, and then you've got yourself an agent.

Money. Replace "tokens" with "money". You just keep throwing money at the loop, and then you've got yourself an agent.

I had a go at this using the on-device models in edge and chrome, phi4-mini and gemini nano, worked surprisingly well for such small models.

https://ryanseddon.com/ai/how-to-build-an-agent-on-device/

Can someone please explain the axes: Oracle, Agent, high safety and low safety?

how does opencode use my claude code subscription instead of making me use api key.

thats one of the things stoppng me from rolling my own. having to use pay per use api.

The “valley of it will take our jobs” is an approaching light in the train tunnel.

I live in the “valley”. I battle depression daily that I had before LLMs.

Using LLMs and false guardrails to watchdog inherently deceitful output is a bad system smell.

I know most are “on it”, and I’ve written a coding agent.

But why is this page designed like some brainwashing repetitive Orwellian mantra?

If it’s perceived that we need that, then we’re having to overcome something, and that something is common sense.

So maybe we’ll happily write our coding agents with the intent to stand on the shoulders of a giant.

But everyone knows we’re building the technological equivalent of a crystal meth empire.

Can someone confirm my understanding of how tool use works behind the scenes? Claude, ChatGPT, etc, through the API offer "tools" and give responses that ask for tool invocations which you then do and send the result back. However, the underlying model is a strictly text based medium, so I'm wondering how exactly the model APIs are turning the model response into these different sort of API responses. I'm assuming there's been a fine-tuning step with lots of examples which put desired tool invocations into some sort of delineated block or something, which the Claude/ChatGPT server understand? Is there any documentation about how this works exactly, and what those internal delineation tokens and such are? How do they ensure that the user text doesn't mess with it and inject "semantic" markers like that?

All these images make it impossibly hard to read... gd scroll simulator

Nitpicking. What the author calls sequence diagrams are not that. They are flowcharts.