AI Research
How to Build Judgement about a New Technology or Business
How I Breakdown New Companies/Solutions to Understand the AI Market
Devansh
3 min read
NEWS & BLOGS
Irys Insights
Research, analysis, and perspectives from the team building the future of legal work.
From our Head of AI
Irys University
Research essays by Devansh, our Head of AI.
Jun 2026 · 8 min read
Auditability Is The Product In Legal AI
Why legal AI needs traceable work product, matter-level audit trails, source-backed answers, and repairable workflows instead of confident black-box output.
Read essay
Jun 2026 · 10 min read
Why Prompt-As-Memory Fails In Legal AI
Why legal AI systems need matter memory, structured work state, source-backed context, and durable review trails instead of prompt summaries.
Read essay
Jun 2026 · 6 min read
What AlphaEvolve Teaches Legal AI About Evaluated Work
DeepMind's AlphaEvolve shows why serious AI needs evaluation, memory, and reviewable state, and why legal AI must translate that lesson into matter-centric work product.
Read essay
Get the latest from Irys.
Legal AI insights, product updates, and research — delivered to your inbox.
67 results
Product Updates
Introducing Workbench
Legal work never lived in one place. Now it can. Workbench, now generally available, reads your matter into an Insights Map, then researches, drafts, redlines, and pulls in the apps you already use.
Christian Brown
5 min read
AI Research
The Cheapest Way to Make Open Weight AI Models Better
We raised a 4B model’s accuracy from 32% to 72% without training or fine-tuning.
Devansh
29 min read
![How Rubric-Based Rewards Could Push AI Beyond Math and Code [Guest]](/_next/image?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!87xa!%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Fhttps%253A%252F%252Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com%252Fpublic%252Fimages%252Fab9b43fb-52d5-40da-995d-5b7cd3f91064_896x896.png&w=3840&q=75)
AI Research
How Rubric-Based Rewards Could Push AI Beyond Math and Code [Guest]
Why the next frontier of reinforcement learning depends on teaching models what “good” means in subjective, open-ended tasks.
Devansh
60 min read
AI Research
The Hidden Geometry Behind AI Reasoning
The geometry behind chain-of-thought, reasoning failures, and the paths models take to an answer.
Devansh
16 min read
AI Research
The AI Industry is Going Through a Massive Correction
What Fable 5, the Space X IPO, and other major events taught us about the space-- June 2026 AI Market Report.
Devansh
45 min read
Product Updates
You didn't get into law to be the glue between your tools
You stopped practicing law and became the glue between your apps. Workbench, now in beta, is an AI teammate that uses your tools, does the work, and brings it back to your case.
Christian Brown
6 min read
AI Research
How to Make Legal AI Truthworthy
How you can use Irys's AI platform to get much better results in your legal work
Devansh
17 min read
AI Research
How to Fix Problems with Your AI Tools
A map to the failure points in Generative AI, and What You should Do to Fix them
Devansh
3 min read
![Will AI Destroy Your Jobs [Livestreams]](/_next/image?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!V7Tt!%2Cw_1456%2Cc_limit%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Fhttps%253A%252F%252Fsubstack-post-media.s3.amazonaws.com%252Fpublic%252Fimages%252F6e291d29-2d0e-4bbe-a16b-42551f161820_485x293.png&w=3840&q=75)
AI Research
Will AI Destroy Your Jobs [Livestreams]
Martin Ford on Why Jobs Break From Consolidation, Not Robots
Devansh
17 min read
AI Research
How Big Banks (and their Friends) Fund the Climate Crisis
How fossil-fuel financing, lobbying, greenwashing, and weak regulation keep climate action stuck.
Devansh
25 min read
AI Research
The 4 Secrets that make GLM 5.2 Special (and what they mean for the Future of AI)
Understanding GLM 5.2 Beyond the Headlines
Devansh
33 min read
AI Research
How to use Agentic Coding Tools like Claude Code or Codes Effectively
A guide to the most powerful (and most misunderstood) AI system on the planet
Devansh
32 min read
Product Updates
The Matter Now Knows Everything
Three updates are live: Workspaces organize your practice, Tasks and Ask Irys are built into every matter, and a redesigned Library connects documents to the tools that handle them.
Irys Team
4 min read
AI Research
How Elon Musk Found is about to Steal from Your 401(k)
Why SpaceX's $1.77 trillion IPO will be force-purchased by your retirement fund , and what you can still do to stop it
Devansh
33 min read
AI Research
How to Deploy AI Projects That Create Business Value
How to Tell If an AI Project Is Actually Worth Shipping
Devansh
3 min read
Industry Insights
Why We Believe the Future of Legal AI Is Stateful
Most legal AI rereads your matter from scratch every time. We built Stateful Swarms to read once, remember, and query cheaply — 39x lower cost and stronger results on Harvey’s benchmark.
Sabih Siddiqi
8 min read
AI Research
Stateful Swarms make AI 100x More Intelligent per Dollar
How Irys beat Harvey AI and Anthropic on the Legal Agent Benchmark (1.7x performance with 39x lower cost)
Devansh
25 min read
AI Research
Token Maxing: The AI Industry is Struggling with Measuring Value
Goodhart's Law is impacting AI across the board.
Devansh
19 min read
AI Research
How to Thrive in the Age of AI
Why You Should Read: James Wang.
Devansh
21 min read
AI Research
The Biggest Opportunity in AI Right Now
How AI is eating itself, and what an Old German Philosopher can teach us about Contrarian Bets
Devansh
35 min read
AI Research
Cerebras: The $56.4 Billion IPO Challenging NVIDIA’s Memory Wall
The physics behind wafer-scale AI chips, why they make LLM decode faster, and why capacity, software, and unit economics may still decide the company’s fate.
Devansh
47 min read
AI Research
How to control your AI Outputs (better than Finetuning)
We’ve been using flat-earth math to navigate warped AI models. Here is the geometric fix.
Devansh
40 min read
AI Research
Why Legal AI Hallucinations Are Three Different Problems, And Most Tools Only Catch One
It takes time to create work that’s clear, independent, and genuinely useful.
Devansh
13 min read
AI Research
ChatGPT vs Gemini vs Claude: The Best LLM Subscription You Should Buy (2026 Edition)
In a World with Claude Code, OpenClaw, and a hundred spinoffs, which AI Subscription Is Worth Paying For in 2026 ?
Devansh
4 min read
Industry Insights
The Real Decision Isn't Harvey vs. Legora vs. Claude.
Harvey starts at $300K. Legora at $30K. For roughly 95% of US law firms, neither is even reachable. The actual choice is Claude DIY vs. a legal-native platform. Here is the math.
Christian Brown, J.D.
8 min read
AI Research
How the Next Generation of AI Models are Going to Completely Change AI Inference
Why Changing from Autoregressive Language Models to Diffusion Models will change AIs hottest problems
Devansh
33 min read
Product Updates
Introducing Agent Mode: Legal Research That Shows Its Work
Most legal AI gives answers. This AI legal research tool shows the investigation behind them.
Sabih Siddiqi
3 min read
AI Research
Google's Gemma 4 will Change How AI Models are Built
Breaking down the architectural decisions Google made — and why edge and server models are built on opposite logic.
Devansh
26 min read
Industry Insights
What Solo Lawyers Should Actually Look for in Legal AI
Legal AI for solo lawyers isn’t the same as enterprise tools. Here’s what actually matters when choosing a legal AI platform.
Alex Silber
4 min read
AI Research
Anthropic's Claude Mythos Launch Is Built on Misinformation
A primary-source investigation for developers and security researchers who want the real story about what the Data says about Mythos
Devansh
39 min read
Industry Insights
What Legora’s $100M ARR Means for Small Firm Lawyers
Legal AI is scaling fast, but not always for small firms. Here’s what BigLaw platforms mean for AI for lawyers.
Alex Silber
3 min read
AI & Law
The Handoff Problem: Why Legal Work Gets Worse Every Time a Case Changes Hands
Every case handoff loses context. Here’s why legal work breaks across transitions and how a legal AI platform fixes it.
Sabih Siddiqi
4 min read
AI Research
AI Isn't a Software Business Anymore
AI Market Report March 2026: Google’s price war, America’s shadow-banking energy play, and why the whole AI stack just fractured into four industries — each with different economics.
Devansh
17 min read
AI & Law
The Family Law Document Problem: Why Legal AI Must Structure the Record
Most law firm AI tools fail before the work begins. Here’s how legal AI actually handles real case records.
Alex Silber
4 min read
AI Research
Come Meet me in SF and Vegas
Saleforce TDX, Google Cloud Next and more..
Devansh
3 min read
Product Updates
Irys Inside Microsoft Word.
Lawyers work in Word. Starting today, the full Irys sidebar lives directly inside Microsoft Word.
Irys Team
3 min read
AI Research
How to use AI for Cutting Edge Research
Featuring insights from Andrew Ng, Terence Tao, Yann LeCunn, and many other top researchers on how they use AI for better gains
Devansh
3 min read
AI & Law
Courts Aren’t Regulating AI. They’re Defining Its Conditions.
Legal AI just changed: courts now care how your AI handles data, not what it produces.
Sabih Siddiqi
8 min read
AI & Law
When a Focused Tool Becomes a Ceiling: Spellbook vs. Irys One
Document-level legal AI helps with drafting, but breaks at deal complexity. Why AI for lawyers must move beyond single-document workflows.
Sabih Siddiqi
12 min read
AI & Law
Legal AI Needs Matter Context, Not Just Chat Sessions
Most legal AI tools rely on chat. But legal reasoning depends on continuity, and continuity depends on matter context. Here’s why chat-based tools fail without it.
Sabih Siddiqi
5 min read
AI Research
Why Ben Horowitz and I Are Investing in Humanity’s Greatest Untapped Asset
From Societal Bottleneck to High-Yield Infrastructure
Devansh
10 min read
AI Research
Why Some Startups Are Easy to Copy While Others Aren’t
What Apple, GitHub Copilot, and legal AI reveal about how incumbents lose paradigm shifts
Devansh
35 min read
![How a Leading Venture Capitalist uses AI Agents [Guest]](/_next/image?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!-M8X!%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Fhttps%253A%252F%252Fsubstack-post-media.s3.amazonaws.com%252Fpublic%252Fimages%252F2c03f2eb-10cb-4fa0-94b3-e79c07f6c9c1_256x256.png&w=3840&q=75)
AI Research
How a Leading Venture Capitalist uses AI Agents [Guest]
A skeptic’s field guide to building personal agents that are powerful, practical, and safe
Devansh
19 min read
AI Research
The Future of On-Device AI
How Liquid AI built the Best Edge AI Model in the World. And what this reveals about the next wave of model design.
Devansh
53 min read
AI Research
How to Prompt Reasoning Models Effectively
Understanding why Reasoning Models are Different from Normal LLMs and How to Prompt them.
Devansh
3 min read
AI Research
How to Diagnose Failures in Large AI Training Clusters
A practical look at how debugging workflows, metrics, and automated runbooks are used to investigate slowdowns and failures in large-scale model training.
Devansh
19 min read
![Earth Scientist Tells You How Your Land Is Lying to You [Livestreams]](/_next/image?url=%2Fimages%2Fblog-11.png&w=3840&q=75&dpl=dpl_2X93Q1HSqgRXSznvVN8rELwCtwPA)
AI Research
Earth Scientist Tells You How Your Land Is Lying to You [Livestreams]
How AI turns weeks of GIS work into 10 seconds of queries
Devansh
7 min read
AI Research
How Long Context Inference Is Rewriting the Future of Transformers
A clear guide to the new architectures battling the transformer’s memory and inference bottlenecks.
Devansh
77 min read
AI Research
AI Market Report: Feb 2026. Ten Frontier Models in 28 Days
How labs differentiate when models converge, what AI coding tools actually change, and why the backlash against AI became a market force
Devansh
3 min read
AI Research
Why We Must All Support Anthropic AIs Stand Against AI-Surveillance and Weapons Systems
How Mass AI Surveillance and AI Weapons Systems Can Threaten Our Freedom
Devansh
42 min read
![Most Important AI Updates of the week. Feb 16th 2026-Feb 22 2026 [Livestreams]](/_next/image?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!qCU8!%2Cw_1456%2Cc_limit%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Fhttps%253A%252F%252Fsubstack-post-media.s3.amazonaws.com%252Fpublic%252Fimages%252F2352a158-b110-4fd2-95a2-0dd5c06ad763_2208x2208.jpeg&w=3840&q=75)
AI Research
Most Important AI Updates of the week. Feb 16th 2026-Feb 22 2026 [Livestreams]
why AI benchmarks are failing and more
Devansh
18 min read
AI Research
The Real Cost of Running AI
From FLOPs to GPUs to KV Cache — What Every Token Actually Costs
Devansh
52 min read
AI Research
How I taught myself to get cutting edge in AI
How to teach yourself AI- for technical folk
Devansh
3 min read
AI Research
How Fractals Can Improve How AI Models Internally Represent Information
A deep dive into hierarchy and why present AI systems are missing it.
Devansh
30 min read
![Why Silicon Valley Got Industrials Wrong [Livestreams]](/_next/image?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!J8-w!%2Cw_1456%2Cc_limit%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Fhttps%253A%252F%252Fsubstack-post-media.s3.amazonaws.com%252Fpublic%252Fimages%252Fb51977d4-15a3-46b6-bbda-05c1a2d3b24b_1608x1848.png&w=3840&q=75)
AI Research
Why Silicon Valley Got Industrials Wrong [Livestreams]
Meet Philipp Wehn, the a16z Founder using AI to improve Heavy Industries.
Devansh
12 min read
![Inside Zhipu: How One of China’s AI Tigers Thinks About Building Models [Livestreams]](/_next/image?url=https%3A%2F%2Fsubstackcdn.com%2Fimage%2Ffetch%2F%24s_!0Df5!%2Cw_1456%2Cc_limit%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Fhttps%253A%252F%252Fsubstack-post-media.s3.amazonaws.com%252Fpublic%252Fimages%252F27a82974-9043-49e2-956b-2cb71e0bf711_2614x2614.jpeg&w=3840&q=75)
AI Research
Inside Zhipu: How One of China’s AI Tigers Thinks About Building Models [Livestreams]
A candid conversation on tradeoffs, scale, and what actually matters.
Devansh
12 min read
AI Research
AI Isn’t Accelerating. It’s Settling.
How verification costs, retention curves, and stateful systems are reshaping where value accumulates
Devansh
3 min read
AI Research
Why Open Source AI is Suddenly Raising Hundreds of Millions of Dollars
Databricks, Sakana AI, Radix, and why Investors are rushing to invest in Open Source AI
Devansh
18 min read
AI Research
Why Meta Keeps Outperforming the Financial Analysts
A deepdive into the Metaverse, Open Source, and how Meta is tackling their biggest challenges.
Devansh
12 min read
![Meet Wafer: The YC Startup Making GPUs 5x More Efficient with AI [Livestreams]](/_next/image?url=%2Fimages%2Fblog-13.png&w=3840&q=75&dpl=dpl_2X93Q1HSqgRXSznvVN8rELwCtwPA)
AI Research
Meet Wafer: The YC Startup Making GPUs 5x More Efficient with AI [Livestreams]
What are Kernels and Why writing them better can improve GPU usage by 5x.
Devansh
9 min read
Press
Irys Prime Launches, Introducing True Legal Intelligence
Irys today announced the general availability of Irys Prime, its matter-aware legal intelligence platform that delivers jurisdiction-grounded retrieval, full audit provenance, and AI-assisted drafting designed to meet professional responsibility standards.
Irys Press
5 min read
AI & Law
Authority Is Not Strategy: How Irys Separates Legal Retrieval from Legal Reasoning
Most legal AI tools conflate finding relevant cases with knowing what to do with them. Irys draws a deliberate architectural boundary between the two. and it changes everything about how lawyers can trust AI in practice.
Sabih Siddiqi
7 min read
Press
The Myth of Legal AI: Why No One is Doing It Right (...Except Us)
In a new analysis, Irys founder Sabih Siddiqi argues that the current legal AI market is built on a fundamental architectural mistake. and explains what it would take to build the category correctly.
Irys Press
6 min read
AI & Law
Beyond Prompts: Why Legal AI Must Be Built Around Matters, Not Conversations
Chat interfaces gave us a taste of what AI could do for legal work. But a conversation is not a matter. The next generation of legal AI has to be organized around how lawyers actually work. and that means building around the matter, not the message.
Sabih Siddiqi
6 min read
Press
Ethical Implications of AI in Law: Ensuring Transparency and Accountability
As bar associations across the country issue updated guidance on AI use in legal practice, Irys outlines how the platform was designed from the ground up to address the competence, confidentiality, and accountability obligations that govern attorney conduct.
Irys Press
4 min read
Industry Insights
Wrappers Aren't Enough. But Neither Is Raw Frontier AI. Law Needs Infrastructure.
The legal AI market has bifurcated into thin wrappers on general models and raw API access for sophisticated teams. Neither is sufficient. Law needs a purpose-built infrastructure layer. and the firms that understand this earliest will have a durable advantage.
Alex Silber
8 min read