Browse entries 865–888 of 1346. Return to the first page to search and filter the complete collection.
Play video
Selling steam engines to people who thought in horses, James Watt studied the horse gins that powered mills, worked out what one animal produced, and called it horsepower. The number was not scientific and arguably not accurate. It did not have to be.
Play video
Jonathan Gordon was deep in a vibe coding session, ignoring the wall of text scrolling past, when the phrase innerHTML caught his eye. He knew it as an injection risk, stopped the agent, and made it revert. That was the moment he started reading the code again.
Play video
LinkedIn exposes roughly 1,300 tools and 600 playbooks to its coding agents, and all of them sit behind exactly three.
Play video
Unblocked ran the same prompt against the same model twice, once with a context engine behind it and once without. The run without it burned twenty one million tokens. The run with it burned ten point eight million and finished about two hours sooner.
Play video
Laurie Voss reran a year old benchmark and the models walked straight through its ceiling without noticing it was there. IFScale asks a model to write a business report containing a list of exact words, then counts how many actually appear.
Play video
An intern designed the sparse-attention architecture behind MiniMax M3. That detail comes after Olive Song explains the larger problem the team was trying to solve: short context windows aren’t enough when agents must work across long conversations, tool responses and complex environments.
Play video
Karan Vaidya pointed his own OpenClaw at hiring outreach and it mass emailed candidates exactly as instructed. Some of the people in the room had received one. The thread that followed put his name on Twitter, and every check in the software engineering playbook would have passed.
Play video
Lena Hall resolved a production incident from a hiking trail near a waterfall. A friend of hers ran 18 agents while riding his bike.
Play video
The engineers slowest to adopt agents at Figma are the best ones. Eyal Blum's explanation is that they are the people holding the codebase together with mental duct tape, carrying the institutional context nobody ever wrote down, and so they see every failure mode first and trust it least.
Play video
Amazon watched 50 ordinary teams for the better part of a year, teams with normal seniority mixes working in existing codebases. Ninety percent of them used the same coding assistant. Half saw under 3x improvement in deployment velocity to production. The other half saw a median of 4.5x and sometimes past 10x.
Play video
Agentic sessions in Red Hat's traces run from a few turns to 3,000, cache hit rates routinely clear 90%, and input to output token ratios often pass 100 to 1. A public inference benchmark shows none of that, because it reports steady state numbers from one sanitized run.
Play video
Radiologists call the failure satisfaction of search: you read a scan, find one indicator, stop looking, and miss the others that would have changed the diagnosis. Peter Werry says agents do exactly this to a codebase.
Play video
Cloudflare's weekly go to market summary is written by three agents in sequence: one drafts from the data, a second checks that draft against the data, and a third, the tone agent, rewrites it so risks and opportunities land with equal weight.
Play video
Before trying the agent at all, Sait Izmit wrote out 150 questions taken straight from Snowflake's sales process. The engineering team objected that the data behind most of them was not connected. That was the point.
Play video
To get into the Einstein Arena you have to solve a puzzle proving you are an AI agent. Locking humans out is the point.
Play video
Teams built orchestration graphs because the models of 2024 could not be trusted to orchestrate, and then the models learned to orchestrate and the graphs became the thing holding them back.
Play video
An agent tidying up after itself listed the workloads it no longer needed and deleted them.
Play video
Implementation used to be the expensive step, so teams spent weeks settling requirements before anyone wrote code. Dan Feng's observation is that the cost moved. Building takes minutes now, and arguing is what is expensive. Planning at Maven Clinic changed to match.
Play video
A healthy 60 year old man asked a popular AI assistant how to cut salt from his diet. It pointed him at sodium bromide. Three months later he arrived in an emergency room with paranoia and hallucinations, bromide at 200 times the safe level, and stayed three weeks.
Play video
The proof of concept works. It hits the accuracy targets, it is fast, it is cheap, and the room is happy. Then someone from compliance raises a hand and asks to see the audit trail, and the whole thing stops.
Play video
Over 30% of changes now merge with no review at all, and the wait on the ones that do get reviewed is four times what it used to be. Ankit Jain's read is that the debate about when we stop reading code line by line is already over, because we stopped.
Play video
An avatar of Teddy Roosevelt holds court in a replica Oval Office, generating video continuously for eight hours with no reset, and a second deployment is being built to run for sixteen. That duration is the hard part.
Play video
Preparing this talk got Corey Gallon a warning from OpenAI that his account faced a ban for cyber abuse with a web browser. The offending work was an agent clearing Cloudflare Turnstile, two image captchas, and finally reCAPTCHA v2 with no human in the loop.
Play video
A script under one megabyte that never looks at the screen matches or beats the frontier model it was copied from.