Videos

Curated talks and discussions on AI agents, developer tools, and platform engineering

Filtering by: Clear filter

No videos found with this tag.

View all videos

September 2026

Dexter Horthy: Why We Stopped Trusting AI to Write the Plan

AI Native Dev Sep 22, 2026 56 min

HumanLayer CEO Dex Horthy on the “YOLO pull request” experiment: 4–5 months of reading the plan instead of the code, a model-designed six-service architecture, and a codebase that became unusable. In the rebuild, co-founder Kyle hand-typed the data architecture for the first two weeks. Dex now treats planning as expected-pain management and argues maintainability has no fast oracle. He backs compounding factory improvements over big-bang rewrites, and says there will probably always be alpha in reviewing something. For now, that's the code.
Spec Driven Development Code Review Software Factories Context Engineering Code Maintainability

Mojo Just Became a Real Programming Language

Source Compiler Sep 6, 2026 12 min

Mojo 1.0 shipped on August 11, 2026, and the headline change isn't a feature — it's a source-stability contract. A unified pointer model puts “unsafe” on the operation rather than the type, interior origins catch references invalidated by container mutation at compile time, and the Python interop hot path got roughly 12x cheaper. Landed two weeks after Qualcomm closed its acquisition of Modular, one week before the compiler itself went Apache 2.0. Includes a sourced accuracy review of the video's claims.
Mojo Programming Languages Systems Programming Modular

I Quantized Qwen3.8-27B. “4-Bit” Doesn't Mean What You Think

Chris Hay Sep 3, 2026 15 min

Chris Hay takes Qwen3.8-27B from ~55 GB down to an 18.8 GB GGUF running in llama.cpp on a laptop — then argues the “4-bit” label hides almost everything interesting. Four bits is a code width, not a storage cost (the block lands at ~4.5 bits/weight, 4.5012 with overhead); some surfaces structurally can't use the block format, others were held at 16 bits by policy. A model doesn't have a precision — it has a precision program, and quantization looks a lot more like compilation.
Quantization Model Representation LLM Internals Local Inference
July 2026

Your Roadmap Is Why You're Losing to AI-Native Teams

NATE B JONES Jul 12, 2026 · 18 min

The gap between Anthropic/OpenAI and everyone else isn't better AI — it's that AI-native teams moved their repeatable coordination out of meetings and documents and into working code that agents can act on. Nate codifies this as 15 “commandments” spanning speed, product, writing, and human culture, and insists they only work adopted all at once as one system — partial adoption just manufactures chaos. Includes the full 15-commandment table up top.
AI Native Operating Model Product Management Culture

AIE World's Fair 2026 — Day 1 Recap: Software Factories & Loopcraft

AI ENGINEER Jul 1, 2026 · Day 1 Recap

Day 1 of AI Engineer World's Fair 2026 gathered ~7,000 attendees under the banner of Software Factories & Loopcraft. Swix opened by framing AI engineering as the craft of stacking loops; the mainstage then moved through knowledge and retrieval (Microsoft), open agentic stacks and new frontier models (OpenAI Codex, Z.ai GLM, Hugging Face × MiniMax), and orchestration at scale (OpenClaw). It closed on a sharp debate about the limits of automation — HumanLayer and Linnet's Labs pushing back on slop and unsafe agents, and Cursor showing where recursive model self-improvement is heading.
Day 1 Recap Software Factories Loopcraft Keynotes AI Engineering

AIE World's Fair 2026 — Day 2 Recap: Autoresearch & Keynotes

AI ENGINEER Jul 2, 2026 · Day 2 Recap

Day 2's mainstage opened on Anthropic's Fable — the new 'mythos' class of models — with Tariq Shihipar's field guide, then turned to verification (Sonar), perception agents (Amazon AGI), and Google DeepMind's Benoit Schillings declaring that 'code is over.' A mid-stream wave of autoresearch and agent-as-a-judge eval talks led into the afternoon keynote block. There, a closing 'build the factories' keynote on human accountability set up Artificial Analysis on the cost of intelligence and Arena on real-world agent evals.
Day 2 Recap Fable Autoresearch Evals Keynotes

AIE World's Fair 2026 — Day 3 Recap: Harness Engineering & Startup Battlefield

AI ENGINEER Jul 3, 2026 · Day 3 Recap

Day 3 of the AI Engineer World's Fair 2026 was themed "harness engineering" — the argument that the bottleneck is no longer the model but everything wired around it. The morning keynote ran from Amplify's State of AI Engineering survey through John Ousterhout's Homa networking protocol, DSPy's case for separating task from model, a Mike Krieger fireside on building with Fable, and Neo4j's ontology-based semantic layer. The afternoon opened with the Great Loops Debate and built to Theo, YC's Garry Tan, and Airtable's Howie Liu before the Startup Battlefield finale, where Comment.io took the $50K grand prize and the four-day, 7,000-attendee conference closed.
Day 3 Recap Harness Engineering Loops Debate Startup Battlefield Keynotes

A Field Guide to Fable: Working With a New Class of Models

AI ENGINEER Jul 2, 2026 Day 2

Anthropic's Tariq Shihipar gives a four-part field guide to Fable, the new "mythos" class of models rolling out later that day. He covers unhobbling Claude to unlock its spiky capability overhang, using Fable to surface your own unknown unknowns, the emotional grief of leaving hand-coding behind, and why he plans to be less reasonable — treating trade-offs as no longer real.
Fable Anthropic Model Behavior Prompting Keynote

In the Land of AI Agents, the Verifiers Are King

AI ENGINEER Jul 2, 2026 Day 2

Sonar CEO Tariq Shaukat makes the case that as coding agents get dramatically more capable, the hard part is no longer generating code but verifying it. He walks through the data on AI slop, tech-debt accumulation, and the diminishing returns of raw velocity, then lays out Sonar's "agent-centric development cycle" (AC/DC) — guide, verify, solve — with zero-trust, multi-layered verification at the center.
Verification Sonar Code Quality Agentic Coding Keynote

Perception Agents: Completing the Loop on Computer Use

AI ENGINEER Jul 2, 2026 Day 2

A member of technical staff at the Amazon AGI Lab argues that clicking and typing were the easy part of computer use — the hard part is reliability in messy, unverifiable knowledge work. The talk introduces "perception agents" that read the rendered screen the way a person does, keep watching to confirm their own output, and act on what you point at. It closes with the open-source release of the first two pieces of a perception-agent harness: a Chrome-extension annotation tool and a design-spec verification tool, demoed live and even driven by audio from a Bee wearable during a meeting.
Perception Agents Computer Use Amazon AGI Lab Verification Keynote

Code Is Over, But There's Plenty To Do: The AI Frontier of Software Engineering

AI ENGINEER Jul 2, 2026 Day 2

Benoit Schillings, VP of Research at Google DeepMind, traces software through three eras — machine limits, human-brain modularity, and now the AI frontier — and argues that superhuman syntax generation means "code is over, but there's plenty to do." With ~80% of new GitHub code now machine-generated, training data is running out, so DeepMind leans on AlphaZero-style self-play to push superhuman coding further. As writing code becomes free, the hard problems shift to design adequacy, active security guardrails, correct-by-construction code, better evals, thinking beyond chains of tokens, and new languages built for models rather than humans.
Google DeepMind Software Engineering Gemini Self-Play Keynote

Keeping the Human in the Loop: Judgment, Accountability, and Owning the Verdict

AI ENGINEER Jul 2, 2026 Day 2

As harnesses, loop engineering, and software factories let agents do more of the work, Aditya Mani argues the scarce resource is judgment backed by evidence. He reframes the engineer's role around what only a human can be answerable for — deciding, verifying, approving, and owning what reaches production. The takeaway: build the factories, keep the lights on, own the verdict.
Human in the Loop Agents Accountability Software Factories Keynote

The Cost of Intelligence

AI ENGINEER Jul 2, 2026 Day 2

Artificial Analysis co-founders Micah Hill-Smith and George Cameron break down the cost of intelligence in mid-2026. Using their Intelligence Index and the new AA Briefcase agentic knowledge-work benchmark, they show a paradox: token prices for a fixed level of intelligence keep falling 5-10x per year, yet single agentic tasks now cost $20 or more. They unpack the four cost drivers — token price, agent turns, token usage, and prompt caching — and close with the chart they call the most important for understanding AI in 2026: intelligence versus cost per task.
Artificial Analysis Benchmarks Cost of Intelligence Agents Keynote

Building Agent Evals in the Real World

AI ENGINEER Jul 2, 2026 Day 2

Wei-Lin Chiang, co-founder and CTO of Arena (LMArena), explains why evaluating agents is fundamentally different from evaluating chatbots. Arena launched an agent mode that has collected over a million real agentic traces, then mines explicit, implicit, and environment signals to build a live leaderboard using a randomized-control-trial methodology. The talk covers the scale of Arena's data, the technical challenges of measuring multi-component agents, and practical advice for anyone building agentic apps.
Agent Evals Arena LMArena Leaderboards Keynote

The Future of Evals: Agent as a Judge

AI ENGINEER Jul 2, 2026 Day 2

Arize co-founder and CPO Aparna Dinakaran argues that as agents grew from prompt-answering to long-horizon, sub-agent systems, classical LLM-as-a-judge evals stopped catching their failures. She introduces "agent as a judge" — adaptive, dynamic analysis for non-deterministic trajectories — and Arize's new Signal agent, a long-running evaluator that reads traces, discovers issue patterns, and can open a PR with a fix. Her take: the future of evals is running all three approaches together.
Evals Agent as a Judge Arize AI LLM Observability Keynote

The 2026 State of AI Engineering Survey

AI ENGINEER Jul 3, 2026 Day 3

Bar of Amplify Partners presents the 2026 State of AI Engineering survey, drawn from 1,048 respondents and run in partnership with Notion and Vercel. She walks through what AI engineers are actually building — modalities, model choice, cost, and agents — and highlights the year's biggest shifts: image generation doubling, cost becoming a first-class constraint, and write-enabled agents tripling. She closes with rapid-fire five-year bets, including a divisive question about AI compute in space.
State of AI Engineering Amplify Partners Survey Agents Keynote

Latency Matters: Why AI Needs Homa, a Clean-Slate Alternative to TCP and RDMA

AI ENGINEER Jul 3, 2026 Day 3

Stanford's John Ousterhout argues that AI networking is shifting from throughput-dominated bulk transfers toward small, latency-sensitive message exchanges in inference and agentic workloads. Legacy protocols like TCP and RDMA suffer high tail latency when small messages mix with large ones. He introduces Homa, a clean-slate datacenter transport that controls congestion from the receiver, prioritizes short messages, and cuts tail latency by an order of magnitude.
Homa Networking Datacenter Tail Latency Keynote

Separating the Task From the Model: Specs, Code, and Evals With DSPy

AI ENGINEER Jul 3, 2026 Day 3

DSPy core contributor Maxime Rivest and lead maintainer Isaac Miller argue for treating AI programs like functions: fix a clear input/output contract, then let the model, prompts, and harness become swappable implementation details. They frame a full task specification as three things — specs (instructions), code (constraints), and evals (what good looks like) — and preview DSPy 4 features like DSPy.flex and qualitative learning for last-mile AI engineering.
DSPy AI Programming Optimization Evals Keynote

Fireside with Mike Krieger: Building With Fable, Delegation & Anthropic Labs

AI ENGINEER Jul 3, 2026 Day 3

Anthropic's Mike Krieger, co-founder of Instagram, talks through the shift from directing every engineering step to describing a goal and letting Fable 'go cook on it.' He shares how he ported a Python labs project to TypeScript over a weekend, how Anthropic works via async 'tag' delegation and Claude Code artifacts, and how Anthropic Labs is organized around rapid persevere-or-pivot bets. He closes on startups, vertical AI, and avoiding burnout.
Fable Anthropic Delegation Anthropic Labs Keynote

Thin Agents on a Smarter Substrate: Ontology-Based Semantic Layers for Agents at Scale

AI ENGINEER Jul 3, 2026 Day 3

Neo4j founder and CEO Emil Eifrem lays out a blueprint for running agents at enterprise scale: instead of thick agents with hand-wired data sources, build thin agents on a shared, ontology-based semantic layer. He describes three pillars — a business-facing ontology, a technical ontology, and runtime execution traces — that together solve the problems of data discovery, trust, duplication, and lack of cross-agent learning.
Neo4j Ontology Knowledge Graphs Agents Keynote

Go Bigger: Rejecting the Old Constraints of Software in the Age of Agents

AI ENGINEER Jul 3, 2026 Day 3

Theo Browne (t3dotgg) argues that as models jump from tool-calling to orchestration, the answer is not to get better but to go bigger. He walks through his own "AI psychosis," the skeuomorphic habits engineers cling to, and the way every tier of software has shifted down — startups becoming side projects, and side projects becoming a single markdown file. His closing challenge: if your idea doesn't feel stupid, it isn't big enough.
Agents Going Bigger Developer Identity Keynote t3.gg

Hiring Employable Agents: Product Form Factors for the Agent Era

AI ENGINEER Jul 3, 2026 Day 3

Airtable founder and CEO Howie Liu closes the World's Fair with a product thinker's map of AI form factors — completions, chatbots, agents, always-on "claws," and orchestrating agents. Using a real landscaping-business customer of HyperAgent, he shows agents that recurse, learn from feedback, hand work off to each other, and pause for human approval. His thesis: the human's job shifts from doing the work to unblocking and orchestrating a fleet of agents, and doing so is becoming table stakes for survival.
Agents Product Form Factors Airtable HyperAgent Keynote

Startup Battlefield Finale & World's Fair Close

AI ENGINEER Jul 3, 2026 Day 3

The World's Fair closes with HyperAgent's Startup Battlefield, the finale of its "Founding 500" competition. Three agent-native finalists pitch for a $100K prize pool before judges Josh, Howie and Theo: Kamad (agentic commodity trade), Comment (a multiplayer markdown editor for people and agents) and Built by Foundry (agents that turn creators into founders). Comment wins the grand prize, and the Replit MC wraps AIE WF 2026.
Startup Battlefield HyperAgent Agent-Native Startups Vertical Agents Conference Close

AIE World's Fair 2026 — Opening Keynote: Software Factories & Loopcraft

AI ENGINEER Jul 1, 2026 10 min

Swix (Shawn Wang) opens the AI Engineer World's Fair 2026 by framing the day's theme around software factories and his essay Loopcraft — the idea that AI engineering is fundamentally about stacking loops and knowing when to move up a loop for scale or down a loop for reliability. He extends the loop metaphor from code to human life and civilization, argues agents are generalizing beyond coding into vertical domains, and celebrates the event's growth to roughly 7,000 attendees. He closes by calling the World's Fair itself the highest loop — "the loop that makes loops," where humans gather to figure out what the next loop is.
AI Engineering Software Factories Loopcraft Agentic Loops Keynote

AIE World's Fair 2026 — Microsoft: On AI and Knowledge

AI ENGINEER Jul 1, 2026 17 min

Pablo Castro (CVP & Distinguished Engineer, Microsoft) frames the AI-and-knowledge problem through three lenses: intrinsic knowledge baked into models, extrinsic knowledge that agents retrieve to ground themselves, and learned knowledge captured by continuously improving agents. He traces the exponential arc from IntelliSense (1996) through ML ranking and GitHub Copilot to today's agents, then demos how Microsoft Foundry, Microsoft IQ, and Azure AI Search deliver company-wide grounding and agentic retrieval. He closes on a real learning loop via Foundry's agent optimizer, which hill-climbs an agent's instructions against an evaluation.
Information Retrieval Agentic Retrieval Microsoft Foundry RAG Knowledge Grounding

AIE World's Fair 2026 — OpenAI: The Codex Keynote

AI ENGINEER Jul 1, 2026 20 min

Alexander Embiricos and Roman Huitt of OpenAI argue that far from replacing engineers, AI is triggering a return to engineering's roots — "AI engineers are eating the world." They frame Codex as a layered, open stack (model, responses API, open-source harness, apps server, app-layer plugins) that OpenAI uses internally exactly as it ships to developers, and preview the GPT 5.6 series across cost-efficient (Terra, Luna), fast (Codex Spark, Cerebras at 750 tok/s), and frontier variants. The talk closes on "value maxing" — extracting real value from agents through cost, speed, and parallel cloud execution rather than "token maxing."
OpenAI Codex GPT 5.6 AI Engineering Open Ecosystem Agentic Coding

AIE World's Fair 2026 — OpenClaw: Better Loops, Not More Terminals

AI ENGINEER Jul 1, 2026 9 min

Peter Steinberger ("the claw father," now at OpenAI) argues that the era of juggling 10+ terminal windows was a false peak of productivity where the human was really acting as scheduler, router, and memory. He describes a shift toward managing a long-lived manager agent that delegates to a team of workers, enabled by three changes: server-side compaction, coordination, and automated triggers. As agents get better, the human bottleneck moves from tokens to compute to attention — so the real work is deciding where to spend it. His closing thesis: "The future is not 20 terminals, it's better loops."
Agent Orchestration Long-Lived Agents Developer Workflow OpenClaw Attention Bottleneck

AIE World's Fair 2026 — Z.ai: GLM 4.2 / 5.2

AI ENGINEER Jul 1, 2026 14 min

Zishan Lee of Z.ai dialed in remotely to introduce GLM 5.2 (and GLM 4.2), the company's latest open-weight models. He explained that GLM stands for General Language Model — a name rooted in a 2021 paper on autoregressive blank-filling — and positioned GLM 5.2's coding and agentic capability as sitting between Opus 4.7 and 4.8, with a new "high" thinking level for better token efficiency. Lee argued the case for open weights (security, on-prem control, fine-tuning, co-designing the future) and closed with a "one more thing": Zcode, Z.ai's own Codex-like coding harness.
Open-Weight Models GLM Z.ai Coding Agents Fine-Tuning

AIE World's Fair 2026 — Hugging Face × MiniMax: Open Models & the M3 Fireside

AI ENGINEER Jul 1, 2026 21 min

Thomas Wolf, co-founder and Chief Science Officer of Hugging Face, opens by mapping the open-model landscape — GLM at #2 on Artificial Analysis and the field of Chinese "AI dragons" (DeepSeek, Moonshot/Kimi, Z/GLM, MiniMax) — then hosts a fireside with MiniMax's Olive on M3. Released in June as the top open-source model at the time, M3 is a compact ~400B total / ~20B active (cited as 428B/23B) model with native multimodality (text, image, video) and a functional 1M-token context powered by MiniMax Sparse Attention (MSA). The conversation traces MiniMax's long-context lineage (M1, MiniMax-01), its from-the-first-step multimodal training, its intern-driven research culture, and its commitment to open source.
Hugging Face MiniMax M3 Open Models Sparse Attention Multimodality

AIE World's Fair 2026 — HumanLayer: Harness Engineering Is Not Enough

AI ENGINEER Jul 1, 2026 19 min

Dex Horthy (HumanLayer) argues that the race to build fully automated "software factories" is producing slop because the limits are not in the harness but in how coding models are trained. He shows that reinforcement learning rewards making tests pass, not maintaining codebase quality, so no amount of loop or harness engineering can fix what is fundamentally a model-training problem. His answer is to "turn the lights back on": use AI-assisted product review, architecture, program design, and vertical slices up front so humans can still read every line and move fast safely.
Context Engineering Software Factories AI Coding Code Maintainability Model Training

AIE World's Fair 2026 — Linnet's Labs: Eric Meyer: Provably Safe Agents: Using Type Systems and Proof-Carrying Code to Tame AI

AI ENGINEER Jul 1, 2026 21 min

Eric Meyer, research scholar at Linnet's Labs, delivers a tongue-in-cheek but technically serious tutorial arguing that AI agents are intrinsically dangerous the moment you give them tool calls, and that alignment baked into model weights is not a real safety guarantee. His thesis: treat safety as a programming-language problem. Instead of letting an agent execute its side-effecting loop directly, have the model emit a program (an expression of type IO) that you can statically analyze, taint-check, and formally prove safe before running. He frames this as a repackaging of 1990s proof-carrying code.
AI Safety Agent Security Type Systems Formal Methods Proof-Carrying Code

AIE World's Fair 2026 — Cursor: Model Behavior: Recursive Model Improvement: How Cursor Trains Composer (and Teaches It to Train Itself)

AI ENGINEER Jul 1, 2026 20 min

Lee Robinson, an ML engineer on model behavior at Cursor, walks through how Cursor trains its Composer models via a two-part flywheel: an outer loop of real-world feedback and A/B tests, and an inner loop of high-quality evals and hard, verifiable training problems. He details reward-hacking discoveries, a novel textual feedback RL method, and a SpaceX/Colossus compute partnership. The thesis: as the top model gets smarter, it can distill derivative models that speed up every part of training — a path toward recursive self-improvement.
Model Training Reinforcement Learning Evals Recursive Self-Improvement Agent Infrastructure
June 2026

The Model Doesn't Unpack Its Memory

Chris Hay Jun 10, 2026 31 min

Chris Hay takes a transformer apart by hand on a locally-running Gemma 3 4B to answer how frozen weights can be a queryable database — and finds that the model packs facts on top of each other but never unpacks them. It has no decoder and doesn't need one: it reads each fact by address, and the address is relation + entity, assembled progressively across layers ~23–27. Ends by hand-building an FFN as a key–value store and writing a brand-new fact into Gemma's memory with no training.
Interpretability Transformers Superposition Mechanistic Gemma

Hermes Agent Use Cases That Are Actually Insane

AI LABS Jun 10, 2026 12 min

AI LABS walks through the automations they run daily on the Hermes agent. The standout: two cost-control flags that govern when LLM tokens get spent — Wake Agent decides at runtime whether to call the model, while No Agent runs jobs with zero LLM involvement but full Hermes context, making most monitoring practically free. Plus second-brain-over-Slack, Gmail lead handling, PRD-based competitor tracking, and social-post drafting.

AI Agents Hermes Automation Cron Cost Control
May 2026

Inside My New Notion Second Brain (Designed for AI)

Better Creating May 17, 2026 26 min

Simon tours his 2026 Notion Life OS rebuild for the agentic era — a new Agent & Skills Hub stores personal agent instructions, sub-agent prompts, knowledge bases, and SOP skills that plug directly into Notion's personal agent. AI meeting notes get a dedicated database with skill-based processing. Design thesis: your second brain should route context to AI tools, not just store notes.
Notion Second Brain AI Agents Personal Productivity PKM
April 2026
March 2026
February 2026
January 2026
December 2025
November 2025
October 2025
September 2025
August 2025
July 2025
June 2025
May 2025
April 2025
March 2025
January 2025
October 2024