6 ms·
HarnessTax: How Much Does the Harness Matter for Coding Agents?
- ed_mercer 9d agoDoes this extend to open models like GLM 5.3? This would mean that simply changing the harness to Pi reduces cost in half?
- roywiggins 9d agoI've found the experience of using Pi with local models feels a lot snappier than both OpenCode or Claude Code.
- joshheitzman 9d agoThe provider's middleware also plays a role. I just completed some benchmarks on my bespoke harness and Kilo Code. There's a chart on my LI post here: https://www.linkedin.com/posts/joshheitzman_i-ran-a-fuller-round-of-benchmarks-on-my-activity-7506040800166723584-WN3v https://www.linkedin.com/posts/joshheitzman_i-ran-a-fuller-r... In the case of DeepSeek-V4-Flash-0731 on deepinfra.com there was little difference when both used high reasoning. In the case of that same model on together.ai there was a substantial difference between the two (high reasoning for both again). When using together.ai with Kilo Code the LLM was having a lot of trouble making successful edits. In some cases that meant a lot tries at using the tools and in others it worked around by running scripts. Meanwhile it used the tools from my harness just fine. I've specifically tried to make my tools easy for all of the open weight LLMs to use correctly. That was inspired by getting some errors from Kilo Code at the beginning of the year telling me that the model was having trouble and I should use a smarter model.
- ebrahimisoheil 9d ago[dead]
- jswelker 9d agoMuch of the extra weight of Claude Code and Codex harness are (ostensibly?) for security and alignment purposes. Whether they are effective is an open question, but leaving those dimensions out and calling it a tax is disingenuous, just turning insecurity into a negative externality. "Why pay the waste disposal tax? Dumping into the ocean is free!" Pi actively omits any sort of guardrails and sandboxing in the name of speed and simplicity, so it is not shocking that it is faster and simpler. Doubling the cost of something in the name of vague security is standard operating procedure for big enterprises, maybe even quite cheap.
- shshsjsj 9d ago[dead]
- roywiggins 9d agoIt's not hard to sandbox Pi without adding anything substantial to the actual size of the system prompt, which is what HarnessTax is mostly measuring. Wrapping it in nono.sh costs approximately zero tokens. (I do think Pi should ship with more than zero builtin sandboxing though)
- seanmcdirmid 9d agoYou can sandbox with tool hooks, at least in antigravity, I think a lot of other harnesses support similar features. Sandboxing via the prompt is just dangerous and unreliable.
- zackify 9d agoI have a couple hundred line TUI that just clones an LXD container and I get all my repos and infra isolated in 5s with pi having isolation to that container. Swap between any instance. Port forward to host all automatic. Funny to me how stuff is described these days. Seems like some people lack the full understanding of all these different systems and really overcomplicate their harness setups or use super bloated all in one tooling.
- 9d ago
- Supermancho 9d agoThe term "harness" here is being overloaded for the term "agent", which is worrying. Putting that aside, there are many factors that matter. The "harness" context, the execution pattern (parallel vs sequential), the ability to delegate to other models, etc. Optimal harnesses use concurrent execution + subagents and are not stuck on one model. Cost and performance are impacted GREATLY by these tactics, regardless of the native agent context (instruction). This kind of single-harness analysis is shallow and misleading, although the finding that "Provider-specific optimization does not guarantee the best pairing" is probably correct, depending on how you measure. It is a starting point.
- corv 9d agoWhile sub-agents could theoretically stack turtles indefinitely, there’s limited room when going up the stack because someone will have to take responsibility beyond “git blame”, and that lands squarely with a person or legal entity.
- visarga 9d agoA big part of that responsibility can be put in code tests. Ensuring good test coverage and quality is how you purchase trust in the work agents do. This also reduces the context problem - a test collection has no recall issues, it just runs every time you call it, the whole battery, checks all the things we could check by code in one fast tool call. For the rest, the things we can't test by code, I use manual testing. A large class of problems are intent divergence, when the model passes tests but it didn't do what I asked. For that I keep a log of all user messages in the project history and review it with agents. This intent alignment is repeated from time to time to catch drift. So I see the "why should I trust the work agent did?" problem as a combination of 1. ensure good testing 2. review intent alignment.
- robinpie 9d agoclaude code feels mildly shitty to use in the way that every other vibe-coded-project-got-out-of-hand project does, which is like, not that bad, but it's fucking ridiculous for a 2 trillion dollar company's main companion product
- topaz0 9d agoIt's pretty great to not use, conversely.
- nativeit 9d agoI’m still baffled as to where these 100x valuations came from. Amazon’s market cap is $2.65T, and I don’t see how any AI company comes within a faint whiff of that sort of value.
- CharlieDigital 9d agoClaude Code (CLI) has been really bad with handling window resizing and layout changes. For long sessions, it will freeze while redrawing. On an M3 Max. With 64 GB or memory. Codex CLI does not have this problem, nor does OpenCode. Every time I split the pane, I'm dumbfounded by how bad this is.
- satnhak 9d agoClaude Code is an absolute dog. They hijack the scrollbar. Keybindings don't work properly on WSL. A jumble sale of slash commands that are almost all useless. Stupid flair like "Boondoggling...". I despise everything about it. Compare this to the Codex CLI that is no fluff, minimal and just does it's job. Open AI models are in a totally different league to Anthropic (in the real world) so no loss really.
- mikert89 9d agoAs the model gets smarter, you need to tell it less
- NewJazz 9d agoAs the black box gets bigger, you need to stop expressing yourself even more.
- malnourish 9d agoPerhaps, but then the value of each thing you tell it goes to.
- big-chungus4 9d agoIt goes the opposite way too. Small models need minimal harness like pi or they get overwhelmed by long system message and tool descriptions
- edgyquant 9d agoNo not really, it’s able to fill it more gaps for sure but the way it fills them is totally random and almost always not what you want it to do
- appbeforelunch 9d ago[dead]
- corv 9d agoMy own findings are in line with this research: Having a coding harness is critical but the differences between them are overstated. Personally, I’ve replaced OpenCode with a thin wrapper around Pydantic-AI as the pythonic analogue to Pi-Agent for headless use via Hermes They’d all do the job - I just prefer to compartmentalize for access control. Keeping the harness’ surface area tiny had the added benefit of preserving my understanding and being able to adapt it to my preferred workflow effortlessly
- ygouzerh 9d agoOne thing I am missing to be able to move out of Claude Code, is the auto mode (and the soft_deny and hard_deny settings that can be tuned), with it's classifier checking the output. It's the killer feature from me personally, often when wanting to troubleshoot for example things like Kubernetes workloads. LLMs are now really good at it, but we doesn't want them to like delete a pod. Other harnesses like Codex have often on static rules, like the allow/deny of claude code, that can filter out based on regex. It's quite good already, but sometimes the model can find a way to write something that wasn't anticipated, or in a convoluted way. After, I guess it's something that can be added in an open-source harness like Pi, and add like this new Jev model or something else equivalent
- corv 9d agoI believe Hermes offers something comparable now. Personally, I’m not convinced that a cascade of LLM judges lets me sleep soundly at night. My agents stay in their sandbox and escalating changes to that env or prod is solely in the (human) operator’s domain. Obviously this is less convenient but I think it’s a necessary tradeoff for security and correctness.
- faangguyindia 9d agoI believe it simply uses Sonnet as a classifier.
- yearolinuxdsktp 9d agoCodex has an “Approve for me” mode (“Guardian” internally) that does the same now.
- x312 9d agoClaude Code/Codex charge the user for their extremely bloated one-size-fits-all system prompts (including safety instructions and other stuff users dont want). In my experience if you're using OpenAI/Claude models and paying API costs, almost every other harness beats Claude Code/Codex in cost.
- spike021 9d agoSay I'm using Claude Code or GPT Codex's harnesses but also sending some queries to the respective Anthropic and OpenAI models via OpenRouter. Do harnesses and therefore sending the queries directly to the LLM providers have caching and other benefits that OpenRouter does not provide? Would I get any of those benefits if I simply proxied any requests to the major providers' harnesses through OpenRouter? Or only if the requests go straight from the harness to the provider's API?
- CaveTech 9d agoYour question is worded kind of confusingly, but all caching is handled on the inference layer, and by all major providers. In short, caching should work as long as you are sending requests to the same model and provider.
- spike021 9d agoApologies. I guess my main question is just is there a benefit to routing directly to a provider through its own harness vs just sending chat requests to those providers using OpenRouter or similar as a proxy? As per your response it sounds like at least caching would happen for any provider regardless of the request's origin.
- lexicalmathical 9d agoIf this is mostly because of the size of the system prompt, then perhaps in long horizon tasks the "tax" will be less obvious.
- shivrajnag 9d ago[dead]
- nojs 9d agoWe really need better harness benchmarks. It seems there's no reliable source that benchmarks the main harnesses against all open source models. I also wish the discussion around Pi did not always use cost/token count as the metric. It's amazingly token efficient, but how does it stack up again opencode and others if you don't care about token count? My experience is that the harness is mainly polish preventing failed tool calls, bad edits, stuff like that, but doesn't make much difference to the overall "intelligence". But that opencode seems slightly more robust against stupid errors than out of the box Pi due to the additional context it forces through every thread.
- sn0n 9d agoA good benchmark would require a decent number of smaller scoped one off tasks to larger multi step refactors, and also one shot full project of simple to complex varieties. In addition to a series of “conversational” ambiguity filled one-liners.
- leemysw 9d ago[flagged]
- calgoo 9d agoI really disliked how opencode works IMO; the harness tries to do to much in my mind. Switching to Pi was a breath of fresh air for me, and I even use hax for some of my local needs where i dont want to have the giant pile of fertilizer that is NPM or PIP installed. The harness becomes more and more important, the smaller the model is as you need to offload context management as well as memory to the harness. The big models basically just need a bash prompt tooling and you let the model manage everything inside its own context.
- mobelkh 9d agowhat does it do too much out if the box? afaik it only has the basic tools and two agents profiles, maybe lsp support is extra?
- 9d ago
- nirmeetimthebes 9d ago[dead]
- Yashjain413 9d agoI think it’s really important, especially when you look at everything the tool does, from the execution loop and context management to feedback. The harness is basically the underlying source of truth. With coding agents, what I’ve noticed is that a simple task can often be handled with a fairly simple harness. But the hidden cost is really around context. One of the more interesting things I’ve seen is that two different harnesses can make a similar number of model calls while consuming a very different amount of context. I think I recently came across a paper comparing Claude Code and Pi that touched on this. More context, more tooling, focused context, simpler loops, all of these can lead to very different costs and performance, even when the number of model calls looks similar.
- snehesht 9d agoI have been using jcode for past two weeks, honestly I feel its much better compared to Opencode.
- lukax 9d agoWhat matters more is that you use the tools that the target model was fine-tuned on. E.g. for editing files with Claude models you should use Edit(file_path, old_string, new_string, replace_all) but with GPT models you should use apply_patch_call(patch) (where patch is a custom patch string with custom grammar). It appears newer models are better at narive harness tool calls and worse at custom tools that look similar to default tools. https://lucumr.pocoo.org/2026/7/4/better-models-worse-tools/ https://lucumr.pocoo.org/2026/7/4/better-models-worse-tools/
- nativeit 9d agoI’m not really a dev, so hefty pinch of salt with this take, but doesn’t this feel like we’re just inventing new “fuzzy” regex with much more required compute?
- shepherdjerred 9d agocomparing LLMs to regex is like the OG dropbox comment (https://news.ycombinator.com/item?id=9224 https://news.ycombinator.com/item?id=9224). I can understand this take 4-5 years ago but I have no idea how that's your position in 2026
- raffael_de 9d ago> inventing new “fuzzy” regex not RegEx ... more like a "fuzzy" DSL.
- sn0n 9d agoDoesn’t that have more to do with the templating of tool-calls and how using them are presented to the models? Or is that just why my model likes to break out of the sandbox, going strait to exec shell command and editing files using python on the cli?
- kouteiheika 9d ago> It appears newer models are better at narive harness tool calls and worse at custom tools that look similar to default tools. This is true. For example, the other day I was doing some testing with DeepSeek-V4-Flash-0731, and I gave it a tool called "EditFile" with "old_content" argument. It ignored the tool definition and called it with "old_string" instead, presumably because it was trained on a bunch of Claude Code traces. So either make sure your tools look nothing like the tools in the popular harnesses, or make them similar.
- sn0n 9d agoIt matters about as much as where you leave the electrician, he’s gonna use what tools he has to get the job done with what he has.
- pyfd 9d ago[dead]
- ebrahimisoheil 9d ago[flagged]
- axionbraid 9d ago[flagged]
- vintagedave 9d ago> Models may perform better with other harnesses than with their own. So it turns out that your Claude models may not need Claude Code… This is interesting. We built our own harness (CodeBot, an agent for Delphi) and it currently uses OpenAI models; we tuned it for specific behaviours and patterns and I find its behaviour better than Codex. Same models underneath.
- ethanprk 9d ago[flagged]
- epolanski 9d agoOkay but...closed source harnesses change multiple times per week, sometimes per day. And they also tune the inference of the models behind it. You're using different tooling every day. Hard to benchmark.
- edgyquant 9d agoA benchmark would be a series of prompts + tests that check their results you theoretically run every day
- Otterly99 9d agoThe most astonishing thing to me is that Pi harness is basically as efficient as the Codex/Claude. I wonder if the same is true for the smaller models in the 9-32B range? I would expect that these models need more steering, but again I was not expecting this result either.
- nichyjt 9d agoit seems like accuracy isn't affected much by the harness. can someone help me understand what's the value or the secret sauce in closed-source harnesses like claude code then?
- itsmeduncan 9d ago[flagged]
- kittikitti 9d agoI find that in many cases, you ARE the harness. Case in point, Terence Tao utilized the simple chat interface to find the Jacobian Conjecture counterexample. This is arguable no harness at all. I also find that many people will disagree with me but most of the time, they just want a button to press that will solve the problem. If that's the work you'd like to accomplish, then the harness matters, but as the article alludes to, it doesn't need to be complex.
- jedbrooke 9d agoI like to say with LLMs the real intelligence was inside us all along. That said, system prompt (which to your point also comes from the human) matters a lot too. For example a coding harness that doesn’t a have a plan mode will keep trying to write underbaked ideas to code instead of working with the user on the plan. That said, a “system prompt” can also be the first prompt you send to the agent in a vanilla chat window, so I do like your point about the user being the real harness
- vb-8448 9d agoThe problem is that Pi, on any third party harness, cannot really compete with codex or cc due to subscriptions.
- mikeocool 9d agoI believe you can use your codex subscription with Pi, and it’s officially endorsed by OpenAI and uses your subscription usage. Though for cc you are correct, using it with your sub draws from extra usage.
- logicchains 9d agoDeepseek harness can if using Deepseek V4.1 Flash from the official API, it's cheap enough to be competitive even without a subscription.
- wyre 9d agoPi supports OpenAI login for ChatGPT subscriptions. Every 3rd party open-source harness I know of supports ChatGPT subscriptions.
- vb-8448 9d agoIsn't this against TOS of OAI?
- wyre 9d agoNo. It’s against Anthropics TOS though. It isn’t explicitly allowed in OAI’s TOS, however they publicly support Pi and OpenCode’s usage of their Oauth, and because Codex is open-source, it means the machinery to support Oauth login is open-source under Apache 2.0
- pllbnk 9d agoAnthropic is the only provider trying to lock in their users by using the dirty tactics: - Prevent subscription usage on third party harnesses - Append "Co-authored by Claude ..." to commit messages
- 8d ago
- grigio 9d agoI also test harness with "Ship Harness Bench", jcode with its browser integration gives the best results
- ekorondy 8d ago[flagged]
- glub 8d agoI came to conclusion that all I really care about from the harness is interoperability. I want to be able to switch providers, models at any point in any session or between sessions in a single action, and I want my transcripts to be in one single format from the beginning till the end of time. I want one place where I define project prompt, one place where I define project agents, MCPs, config. And all of this immediately rules out any lab-provided harness.
- abhinavssram 6d agoSo did you build your own thats fit this use case ?
- glub 6d agoI did, initially. Then working on it distracted me from my actual work. oh-my-pi then matched most closely to what I had built. I've been running it for the past 2 months with no complaints. Claude implementation is technically ToS breaking, but my 3 accounts are fine so far.