Videos — All · 24 cards

Why Waiting for the Next AI Model Can Pay Off - John Stecher, Blackstone & Scott Worland, Norm AI

20:453 min read

Strategic timing of AI adoption combined with robust data management and rapid prototyping allows organizations to replace obsolete software systems efficiently as model capabilities scale.

How We Took an AI Agent From 41% to 97% Accurate - Milind Sunil Shah, Brainbase Labs

13:233 min read

Refining instructions, rules, and ontologies iteratively across eleven versions raised a back-office pricing agent from 41% to 97% accuracy while cutting costs in half.

Our Smaller Model Beat Opus on Finance Tasks - Alexander Shevchenko, Ramp

14:152 min read

Ramp Labs improves financial agent workflows and reduces operational costs by building specialist post-trained models, offline KV cache cartridges, and large-scale agentic swarms.

AI Labs Will Trap You or Replace You - Dillon Rolnick, Nous Research

15:133 min read

As AI models commoditize and labs attempt to capture trillion-dollar infrastructure investments through proprietary lock-in or direct vertical replacement, organizations must preserve their autonomy by building on open agent foundations like Hermes Agent.

Robots Will Be Trained on Video Games - Pim de Witte, General Intuition

18:512 min read

Training vision models on billions of video game action tokens creates general agents that can control multiple robot embodiments and computer interfaces in real time.

Don't Approve the AI Model. Approve the App - Doneyli De Jesus, Langfuse

19:203 min read

Deploying AI applications in regulated industries requires building custom golden datasets, gating the application rather than the model, and incorporating deterministic tools alongside domain expert oversight.

Why Datasite Stopped Building Its Own Parsing - Adit Abraham, Reducto & Raj Bakhru, Datasite

19:492 min read

Outsourcing document ingestion and parsing to specialized vendors allows engineering teams to focus on core revenue-generating product features like GraphRAG and AI agents.

Building the AI Agent Is Easy. Proving It's Safe Isn't - Jonathan Osborne, NatWest

18:013 min read

Deploying customer-facing AI agents in a regulated banking environment requires moving beyond simple accuracy scores to establish rigorous continuous evaluation, multi-layered routing funnels, and calibrated LLM judges.

Why AI Makes You 100x Faster but Not Your Company - Mitchell Troyanovsky, Basis

19:133 min read

Production agents require opinionated design and process supervision to bridge the enterprise productivity gap instead of relying solely on individual coding tools.

Without Caching, Your AI Agent Costs 10x More - George Cameron, Artificial Analysis

19:433 min read

Prompt caching and model selection drive agent costs, where lacking cache hits multiplies task expenses tenfold and different models span multiple orders of magnitude in price.

Make Your AI Agents Predictable With Graph Engineering - Zachary Blumenfeld, Neo4j

20:362 min read

Designing agent workflows as explicit graph structures and compiling procedural skills into deterministic formats improves task performance, cuts execution time by half, and significantly reduces operational costs.

Your AI Agents Shouldn't Care Where They Run - Nikhil Harithas, Factory

17:133 min read

Open Compute abstracts the underlying infrastructure lifecycle for software engineering agents, enabling secure deployments directly inside customer Virtual Private Clouds across diverse compute layers.

Better Search Took This AI Agent From 36% to 98% Accuracy - Aamir Shakir, Mixedbread

16:303 min read

Switching from a traditional Frankenstein RAG stack to Mixedbread's latent interaction and multimodal search model increased a hedge fund's agent answer accuracy from 36% to 98% while cutting LLM token costs by 3x.

You Can't Tell What's Real Anymore. Here's What Comes Next - Cristóbal Valenzuela, Runway

16:512 min read

Scaling next-frame prediction turns video models into interactive simulation engines that eliminate the need for traditional 3D engines like Unity or Unreal.

Your Company Should Train Its Own AI Models - Will Brown, Prime Intellect

20:163 min read

Organizations can capture compounding institutional advantages by training custom open-source AI models on internal workflows and data rather than relying entirely on frontier lab models.

A Valid Key Let an AI Agent Delete Production - Sagar Batchu, Speakeasy

20:403 min read

Unchecked authority rather than model error causes agent failures, making a dedicated control layer that enforces identity, policy, and audit logging mandatory for safe AI deployment.

AI Writes Code Faster Than Your Team Can Review It - Andrew Norrish, Sourcegraph

18:573 min read

AI-driven development increases developer velocity and code volume, but shifts the bottleneck to human code review, resulting in more production incidents and increased SRE toil.

Your AI Pilots Won't Move the P&L. Here's What Will - Adam Nahari, Apollo

15:523 min read

Moving the P&L with AI requires investing five units of change management and end-to-end workflow redesign for every one unit of engineering.

I Don't Fear AI Slop. I Fear Untold Stories - Zack Proser, WorkOS

18:342 min read

Building automated production factories infused with personal craft and domain expertise enables colleagues and children to share untold stories with asymmetric leverage.

I Made a Fruit Fly Review My Code - Nick Nisi, WorkOS

16:372 min read

Automated code review using a simulated fruit fly brain and AI models demonstrates that routine pull request checks can be offloaded, shifting human responsibilities from line-by-line inspection to verifying business logic and context.

Why We Rebuilt Microsoft Office From Scratch for AI Agents - Yijia Liang, Compound AI

15:453 min read

Rebuilding Office suite tools from scratch as headless, AI-native engines enables financial firms to delegate complex, multi-week deal workflows to autonomous agents.

Most of Your Docs Readers Are Now AI Agents - Hahnbee Lee, Mintlify & Lincoln Murr, Coinbase

18:523 min read

AI agents have rapidly become the primary consumers of developer documentation and internet services, driving companies to prioritize lightweight APIs and autonomous microtransactions over traditional human frontend design.

If Your AI Scores 100%, Your Eval Is Wrong - Ramakrishnan Lokanathan, eslams.com

18:453 min read

Effective AI evaluations for high-stakes production applications require testing entire workflows and continuously increasing test difficulty rather than measuring isolated final answers.