Skip to main content
Original: Swyx · 05/03/2026

​
Summary

The article discusses the ongoing debate in Harness Engineering regarding the balance between model capabilities and the importance of the harness that connects them.

​
Key Insights

“The Harness is the Product: Every production agent converges on this core loop.” — Discussing the significance of harness engineering in AI systems.
“The biggest barrier to getting value from AI is your own ability to context and workflow engineer the models.” — Highlighting the importance of effective engineering in AI applications.
“Harness Engineering has real value.” — Concluding thoughts on the evolving perspective of harness engineering in the AI industry.

​
Topics

​
Full Article

A common debate in my finance days was about the value of the human vs the value of the seat: if a trader made 3minprofits,howmuchofitwasbecauseofherskills,andhowmuchwasbecauseoftheposition/institution/brandsheisin,andanygenerallycompetenthumancouldhavemadethesameresults?ThesamedebateiscurrentlyraginginHarnessEngineering,thesystemssubsetofAgentEngineering,andthemainjobofAgentLabs.ThecentraltensionisbetweenBigModelandBigHarness.[AnAIframeworkfounderyouallknow]onceconfidedinmeatanOpenAIevent:Imnotevensuretheseguyswantmetoexist.Aside:letsdefineHarnessIneveryengineeringdiscipline,aharnessisthesamething:thelayerthatconnects,protects,andorchestratescomponentswithoutdoingtheworkitself.And,talkingwiththeBigModelguys,youreallyseeit:EverypodcastwithBorisChernyandCatWuemphasizeshowminimaltheharnessofClaudeCodeis,meaningtheirjobismostlylettingthemodelexpressitsfullpowerinthewaythatonlythemodelmakerknowsbest:Boris:Iwouldsayliketheresnothingthatsecretinthesource.AndobviouslyitsallJavaScript,soyoucanjustdecompileit.Compilationsoutthere.Itsveryinteresting.Yeah.Andgenerallyourapproachis,youknow,allthesecretsauce,itsallinthemodel.Andthisisthethinnestpossiblewrapperoverthemodel.Weliterallycouldnotbuildanythingmoreminimal.Thisisthemostminimalthing.Cat[01:09:21]:ItisverymuchthesimplestthingIthinkbydesign.Boris[01:09:25]:Soitsgotsimpler.Itgotsimpler.Itdoesntgomorecomplex.Weverewrittenitfromscratchprobablyeverythreeweeks,fourweeksorsomething.Anditjustlikeallthe,itslikeashipofTheseus,right?Likeeverypiecekeepsgettingswappedoutandjustcausequadissogoodatwritingitsowncode.OpenAIsownpieceonHarnessEngineering(withupcomingguestRyanLopopoloontheCodexteam)emphasizeshowsimpleitistostart.Ofcourse,withtheexecuhireofOpenClaw,OpenAIarenowbiginvestorsoftheworldsmostsuccessfulopensourceharness.NoamBrown:beforethereasoningmodelsemerged,therewaslikeallofthisworkthatwentintoengineeringagenticsystemsthatlikemadealotofcallstoGPT−4oorlikethesenon−reasoningmodelstogetreasoningbehavior.Andthenitturnsoutwejustcreatedreasoningmodelsandthey,youdontneedthiscomplexbehavior.Infact,inmanyways,itmakesitworse.Likeyoujustgivethereasoningmodelthesamequestionwithoutanysortofscaffoldinganditjustdoesit.Andsopeoplearebuildingscaffoldingontopofthereasoningmodelsrightnow.ButIthinkinmanyways,thosescaffoldswillalsojustbereplacedbythereasoningmodelsandmodelsingeneralbecomingmorecapable.Andsimilarly,Ithinkthingslikemodelrouters,wevesaidprettyopenlythatwewanttomovetoaworldwherethereisasingleunifiedmodel.Andinthatworld,youshouldntneedarouterontopofthemodel.METRsayingClaudeCodeandCodexdontbeatabasicscaffold:ScaleAIsSWE−AtlasisfindsthatOpus4.6does2.5pointsbetterinClaudeCodethaninthegenericSWE−Agent,butthereverseforGPT5.2,makingtheharnessyouchooseessentiallynoisewithinthemarginoferror:Andyet.TheBigHarnessguysdisagree:TheHarnessistheProduct:Everyproductionagentconvergesonthiscoreloop:while(modelreturnstoolcalls):executetoolcaptureresultappendtocontextcallmodelagainThatisit.TheentirearchitectureofClaudeCode,Cursorsagent,andManusfitsinsidethatloop.JerryLiu:TheModelHarnessisEverythingthebiggestbarriertogettingvaluefromAIisyourownabilitytocontextandworkflowengineerthemodels.Thisis∗especially∗truethemorehorizontalthetoolthatyoureusing.Improving15LLMsatCodinginOneAfternoon.OnlytheHarnessChangedshowsdramaticimprovementsineverymodelwhenyouoptimizetheharness(Pi)ObviouslyBigHarnessguysaretryingtosellyoutheirHarness,BigModelguysaretryingtosellyoutheirModel.TheML/AIindustryhasalwayshadsomeformofmilquetoastcompoundAIdebatethattellsyoubotharevaluable.Butperhapsthetimesarechanging.OnLatentSpacewevebeenvery,veryrespectfuloftheBitterLesson,butincreasinglyastheAgentLabsthesishasplayedout(withCursornowvaluedat3m in profits, how much of it was because of her skills, and how much was because of the position/institution/brand she is in, and any generally competent human could have made the same results?The same debate is currently raging in Harness Engineering, the systems subset of Agent Engineering, and the main job of Agent Labs. The central tension is between Big Model and Big Harness. [An AI framework founder you all know] once confided in me at an OpenAI event: Im not even sure these guys want me to exist. Aside: lets define Harness In every engineering discipline, a harness is the same thing: the layer that connects, protects, and orchestrates components without doing the work itself.And, talking with the Big Model guys, you really see it:Every podcast with Boris Cherny and Cat Wu emphasizes how minimal the harness of Claude Code is, meaning their job is mostly letting the model express its full power in the way that only the model maker knows best:Boris: I would say like theres nothing that secret in the source. And obviously its all JavaScript, so you can just decompile it. Compilations out there. Its very interesting. Yeah. And generally our approach is, you know, all the secret sauce, its all in the model. And this is the thinnest possible wrapper over the model. We literally could not build anything more minimal. This is the most minimal thing.Cat [01:09:21]: It is very much the simplest thing I think by design.Boris [01:09:25]: So its got simpler. It got simpler. It doesnt go more complex. Weve rewritten it from scratch probably every three weeks, four weeks or something. And it just like all the, its like a ship of Theseus, right? Like every piece keeps getting swapped out and just cause quad is so good at writing its own code.OpenAIs own piece on Harness Engineering (with upcoming guest Ryan Lopopolo on the Codex team) emphasizes how simple it is to start. Of course, with the execuhire of OpenClaw, OpenAI are now big investors of the worlds most successful open source harness.Noam Brown: before the reasoning models emerged, there was like all of this work that went into engineering agentic systems that like made a lot of calls to GPT-4o or like these non-reasoning models to get reasoning behavior. And then it turns out we just created reasoning models and they, you dont need this complex behavior. In fact, in many ways, it makes it worse. Like you just give the reasoning model the same question without any sort of scaffolding and it just does it. And so people are building scaffolding on top of the reasoning models right now. But I think in many ways, those scaffolds will also just be replaced by the reasoning models and models in general becoming more capable. And similarly, I think things like model routers, weve said pretty openly that we want to move to a world where there is a single unified model. And in that world, you shouldnt need a router on top of the model.METR saying Claude Code and Codex dont beat a basic scaffold:Scale AIs SWE-Atlas is finds that Opus 4.6 does 2.5 points better in Claude Code than in the generic SWE-Agent, but the reverse for GPT 5.2, making the harness you choose essentially noise within the margin of error:And yet. The Big Harness guys disagree:The Harness is the Product: Every production agent converges on this core loop:while (model returns tool calls): execute tool capture result append to context call model againThat is it. The entire architecture of Claude Code, Cursors agent, and Manus fits inside that loop. Jerry Liu: The Model Harness is Everything the biggest barrier to getting value from AI is your own ability to context and workflow engineer the models. This is *especially* true the more horizontal the tool that youre using.Improving 15 LLMs at Coding in One Afternoon. Only the Harness Changed shows dramatic improvements in every model when you optimize the harness (Pi)Obviously Big Harness guys are trying to sell you their Harness, Big Model guys are trying to sell you their Model. The ML/AI industry has always had some form of milquetoast compound AI debate that tells you both are valuable. But perhaps the times are changing.On Latent Space weve been very, very respectful of the Bitter Lesson, but increasingly as the Agent Labs thesis has played out (with Cursor now valued at 50B), we are acknowledging that Harness Engineering has real value. AIE Europe now has the worlds first Harness Engineering track, and if you are keen on this debate, you should join.AI News for 3/3/2026-3/4/2026. We checked 12 subreddits, 544 Twitters and 24 Discords (264 channels, and 14242 messages) for you. Estimated reading time saved (at 200wpm): 1397 minutes. AINews website lets you search all past issues. As a reminder, AINews is now a section of Latent Space. You can opt in/out of email frequencies!AI Twitter RecapFrontier model shipping: Gemini 3.1 Flash-Lite, GPT-5.4 rumors, and agent-first product positioningGemini 3.1 Flash-Lite positioning (speed/):DemisHassabisteasedGemini3.1Flash−Liteasincrediblyfastandcost−efficientforitsperformanceclearlyframingthemodellinearoundlatencyandcostpercapabilityratherthanrawfrontierscores(tweet).RelatedproductchatterhighlightsNotebookLMasafavoriteAItool(tweet)andamajornewNotebookLMStudiofeature:CinematicVideoOverviewsthatgeneratebespoke,immersivevideosfromusersourcesforUltrausers(tweet).GPT−5.4leaknarrative(TheInformation):MultipletweetsamplifyareportthatGPT−5.4iscomingwitha 1Mtokencontextwindowandanewextremereasoningmodethatcanthinkforhours,targetinglong−horizonagenticworkflowsandlowercomplex−taskerrorrates(tweet,tweet,tweet).TheresalsospeculationthatOpenAIisshiftingtomorefrequent(monthly)modelupdates(tweet).Separately,onearenawatcherclaimsGPT−5.4landedinthearena,implyinganimminentreleasewindow(tweet).TreatallofthisasunconfirmedunlesscorroboratedbyOpenAI.Claudeasagentbehaviorleader,notjustcoding:NatLambertarguesthediscussionshouldshiftfromAnthropicgoingall−inoncodetotheirleadongeneralagentbehavior,implyingcodingcapabilitywillcommoditizebutagentrobustnesswillnot(tweet).MathArenaevaluationaddsadatapoint:ClaudeOpus4.6isstrongoverallbutweakonvisualmathematics,andcostlytoevaluate(claimed ): Demis Hassabis teased Gemini 3.1 Flash-Lite as incredibly fast and cost-efficient for its performanceclearly framing the model line around latency and cost per capability rather than raw frontier scores (tweet). Related product chatter highlights NotebookLM as a favorite AI tool (tweet) and a major new NotebookLM Studio feature: Cinematic Video Overviews that generate bespoke, immersive videos from user sources for Ultra users (tweet).GPT-5.4 leak narrative (The Information): Multiple tweets amplify a report that GPT-5.4 is coming with a ~1M token context window and a new extreme reasoning mode that can think for hours, targeting long-horizon agentic workflows and lower complex-task error rates (tweet, tweet, tweet). Theres also speculation that OpenAI is shifting to more frequent (monthly) model updates (tweet). Separately, one arena watcher claims GPT-5.4 landed in the arena, implying an imminent release window (tweet). Treat all of this as unconfirmed unless corroborated by OpenAI.Claude as agent behavior leader, not just coding: Nat Lambert argues the discussion should shift from Anthropic going all-in on code to their lead on general agent behavior, implying coding capability will commoditize but agent robustness will not (tweet). MathArena evaluation adds a datapoint: Claude Opus 4.6 is strong overall but weak on visual mathematics, and costly to evaluate (claimed ~8k) (tweet). Read more

[AINews] The high-return activity of raising your aspirations for LLMs

Swyx · explanation · 72% similar

How I think about Codex

Simon Willison · explanation · 69% similar

[AINews] Sam Altman's AI Combinator

Swyx · explanation · 69% similar
Originally published at https://www.latent.space/p/ainews-is-harness-engineering-real.