Harnesses in AI: A Deep Dive — Tejas Kumar, IBM
298K views · May 17, 2026 · Science & Technology
Comments · 357
@nitingaur1707 · 4 months ago
That's how the tech talk should be delivered. Kudos man!
233
@azafromkaza · 4 months ago (edited)
The main thesis is: extract as much deterministic verifiable actions as possible and provide them as rule-based pieces of code helpers for efficient token usage. When exhausted all rule-based harnessing possibilities, move to tweaking your system prompt. And that is an example of good harnessing. <br>Did i get it right?
246
@suhagmanjeet · 1 month ago
Prompt engineering, Context engineering and now Harness engineering...world trying one thing after another to have deterministic output from a probabilistic system/LLM.
22
@jayhu6075 · 4 months ago (edited)
What a great talk about Harness. I think one of the best talks in AI engineering I’ve seen lately. <br>Tejas Kumar simplified the topic so well — hopefully we get more content from him. Many thnx.
94
@mariuszsiera · 2 months ago
Funny, first we started to use dynamic tools invocation, forgetting about programming. Now programming is back just baked in the agent loop for the specific use case..
7
@exp24816 · 3 months ago
A rare high quality talk that is actually useful and grounded in real world. Thank you so much
5
@gadielshepherd6780 · 4 months ago
The line that stands out most to me is about "failing correctly". Can take a lot out of just that, and not just in the context of software development. Great talk
24
@AninditaBhowmik-m7b · 5 days ago
The best example picked up to explain the concept to a wide range of tech professionals seamlessly!
@naturelstoic · 3 months ago
Had the chance to meet Tejas live at QCon in London, not too long ago. What a kind person! I was genuinely impressed.<br><br><b>That's right, I believe it's rare for an engineer of his stature</b>
15
@ntone.s · 4 months ago
Bro got the energy to explain this so good! Thanks for the talk!
23
@kickinon · 1 month ago
I wish I had teachers like Tejas, very engaging.
1
@OlafFinkbeiner · 4 months ago
Thanks for this talk so good and so IMPORTANT. Great model bad harness = shit. OK model plus great harness = ok experience
68
Up next
When to Build Your Own Agent Harness | Harrison Chase, LangChain
Sequoia Capital · 95K views
Jev for Claude is INSANE (Every Jev Concept Explained)
Simon Scrapes · 91K views
Django Framework Tutorial for Beginners | Session 3 | Learn with Agentic GenAI | Edso Learn
Edso Learn · 9 views
"Software Fundamentals Matter More Than Ever" — Matt Pocock
AI Engineer and Matt Pocock · 1.3M views
AI Model vs Agentic Harness: What Actually Drives AI
IBM Technology · 144K views
Context Rot Is a Harness Problem
Decoding AI · 8.6K views
Princeton Just Proved AI Can’t Take Your Job
Brendan Dell · 255K views
Jev CEO: I made ChatGPT, now I'm building what's next
AI Engineer · 471K views
No Vibes Allowed: Solving Hard Problems in Complex Codebases – Dex Horthy, HumanLayer
AI Engineer · 648K views
Building pi in a World of Slop — Mario Zechner
AI Engineer · 552K views
Agent Harness explained in 8min..
Caleb Writes Code · 610K views
What AI Agent Skills Are and How They Work
IBM Technology · 472K views
The Death of the Code Review: What the Data Actually Says — Laurie Voss, Arize AI
AI Engineer · 15K views
Ex-NASA dev reveals his Agentic Engineering Workflow
David Ondrej · 327K views
Anthropic's Chloe Lubinski explains how AI works (in 14 minutes)
Alliance for Responsible Citizenship · 2.6M views
10 Levels of Jev For Agentic Engineers
IndyDevDan · 80K views
Harness Engineering Deep Dive
Rajistics - data science, AI, and machine learning · 9.8K views
L8 Principal's Agentic Engineering Workflow
Kun Chen · 1M views
Bill Gates: A.I. ‘Makes Nuclear Weapons Look Like Nothing’ | The Ezra Klein Show
The Ezra Klein Show and 2 more · 605K views
The Multi-Agent Architecture That Actually Ships — Luke Alvoeiro, Factory
AI Engineer · 262K views
Loop Engineering explained in 8min..
Caleb Writes Code · 215K views
Yann LeCun's $1B Bet Against LLMs [Part 1]
Welch Labs · 1.2M views
Prompting Is Dead in 6 Months. Andrew Ng, Stanford
philia · 224K views
How to Ship Complex Features 10x Faster with AI Agents | Dex Horthy (HumanLayer)
AI Tinkerers · 24K views
Matt Pocock’s Agentic Engineering Workflow (just copy him)
David Ondrej and Matt Pocock · 446K views
Harness Engineering Explained in 22 Minutes
Shaw Talebi · 42K views