List from buildsoftwarewithai.com
AI for Developers
Tools, techniques and ideas for building software with AI.
Latest podcasts
Newest first
Chat GPT's Creator Explains The Shocking Truth About How AI Understands the World | Ilya Sutskever
Craig Smith sits down with Ilya Sutskever - Co-Founder and Chief Scientist at Safe Superintelligence Inc. & former chief scientist at OpenAI and one of the primary minds behind GPT-3, GPT-4, and the deep learning revolution that preceded it - for a conversation that covers the intellectual history of deep learning and the hardest current questions simultaneously. Ilya’s most important argument is one that most public discourse gets wrong: the critique that LLMs “just learn statistical patterns” misses what prediction at scale actually achieves.
Content Marketing World Recap: You Can Market to Bots Without Sounding Like One
I’m fresh out of the Content Marketing World 2026 conference, and it’s time to debrief. To write this blog post, I could have put my copious notes into Claude, asked it to synthesize the best points, and “write an 800-word blog in my voice.” Instead, I stopped short.
Physical vs. Cloud Servers for Building an Intelligent Home AI
• Every home AI needs a server somewhere. The only question is whether it sits in your house or in someone else's data center. • Cloud servers win on raw compute, elastic scaling, storage, remote access, and low upfront cost. • Physical servers win on privacy, control, offline operation, and response time. • Buyers of a home AI care most about five things: data privacy, control, power efficiency, response speed, and working with or without the internet. A local server takes four of the five.
AI Agents Don't Need User Accounts. They Need Identities
AI agents are becoming autonomous software actors with access to real systems and sensitive data. This article explores why traditional user accounts are not enough and introduces an Agent Identity Lifecycle covering identity, credentials, authorisation, delegation, auditability, revocation, and retirement.Read All
Why a Five-Cent Earnings Beat Isn't Always What It Seems
A five-cent earnings beat looks simple until you ask whether the reported result and the analysts’ estimates measure the same thing. I use a language model to read the meaning buried in analyst notes, but I keep the final decision about which estimates belong in the consensus outside the model.Read All
AI, Robot Cruelty, and the Human Instinct to Dominate
We're creating intelligent machines that may surpass human capabilities, but our instinct is to dominate and control them, raising questions about empathy, cruelty, and our relationship with power. Our actions towards machines, even if they're just inanimate objects, reveal our true nature and the kind of world we want to create.Read All
ThursdAI - Oct 8 - OpenAI drops 722 math papers, Haiku 5.5 hits 10 cents & more
From CoreWeave: GPT-6 is free for everyone, open weights hit 1T params, an agent leaked its owner’s bank balance to Slack, Peter broke Arena’s 200M raise live, and Maxime Labonne on decision models.
Google bid $10 million for an airline’s old work emails to train AI. Much of what made that work matter never showed up in an email
Google wants to learn from the record of our working lives. We should take a closer look ourselves, because understanding what our work accomplishes could change what we attempt next.
The Claim Was Never Verified, And Yet, The Report Looked Official
An AI-assisted intelligence assessment about a Chinese vessel was wrong, yet it acquired enough institutional authority to help set a military operation in motion before human review caught the error. The problem is not simply that generative AI can hallucinate. It is that uncertain conclusions can inherit credibility from classified sources, official language, and formal presentation without actually being verified.
Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
It’s hard to believe that Periodic was only launched last September: One year later, it is considered one of the pre-eminent AI scientist labs, with dizzying talent density and astonishing progress in the autonomous lab buildout: Most people are familiar with the standard credentials of Liam and Dogus, but we found an incredible “talent slope” while learning more about Periodic, where each successive employee seems more impressive than the last:
Inside a Game Where AI Agents Never Sleep
What happens to an MMORPG when half the players never log off? Issac Lee, who runs AI and blockchain experiments at NEXUS, has been testing that idea. His team is filling game worlds with agent players instead of scripted NPCs, so a new title feels alive from day one.We get into why agent-generated content could be the next step after studio-made and user-generated games.
The Investors Who Bought Anthropic Before the World Understood Its Value
Anthropic had two sets of early investors. One funded it before Claude existed. The other bought in after the company was already well known. In 2021, Jaan Tallinn led a $124M Series A at about a $623M valuation, joined by Dustin Moskovitz, James McClave and Eric Schmidt. There was no product yet, so they were backing the founders and the safety research. In 2024, the FTX bankruptcy estate sold its stake for about $1.3B, against roughly $500M invested, at an $18B valuation. Craig Falls bought 666,667 shares for about $20M.
AGI and ASI are levels. RSI is a loop
AGI, ASI and RSI get used as if they were one idea at three volumes. They aren’t. AGI and ASI name levels of capability. RSI names a process, a way a system might move from one level to the next. Once you separate the two kinds of term, most AI headlines get easier to read.
SE 31: Kartik Sakthivel, CIO at John Hancock | Live from Ai4 2026
Send us Fan Mail Kartik Sakthivel is the Chief Information Officer at John Hancock, the life insurer that turned 164 years old in April 2026 and is part of Manulife.
Narrative intelligence and the human advantage
As AI reshapes how we work, create, and think about our roles, staying adaptable means understanding not just the technology, but our own stories and identities. Reed Frerichs joins Daniel and Chris to explore narrative intelligence, storytelling, leadership, and personal growth in an AI-driven world. They discuss how entrepreneurs and leaders can use AI tools to navigate changing roles, embrace uncertainty, and develop emotional intelligence along with Reed's “own, author, act” framework.
AI Can Remember Your Chats. Remembering Your Life Is Much Harder
For AI to become truly useful in everyday life, it needs memory that goes beyond chat history. Building that memory means solving four key challenges: deciding what to remember, retrieving the right experience, connecting moments over time, and learning from corrections as people and circumstances change. At Looki, we’re exploring how visual evidence, supported by audio, chat, location, and other context, can create AI memory that is grounded in what actually happened—and remains understandable, verifiable, and correctable.Read All
The Eyes of Intelligence: Why Computer Vision Will Define the AI Century
For three years we've measured AI progress in text such as tokens, context windows, benchmark scores. That framing treats intelligence as a language problem and everything else as a downstream app. I think that's backwards. The capability that decides whether AI stays on screens or steps into the world is perception: a machine looking at the physical world and being right about what it sees.
I Wrote a War Fought With AI Weapons. How Close Is Fiction to Reality?
AI is already changing warfare. Drawing on the ideas behind An Unlikely War, this article explores autonomous weapons, human control and what happens when military decisions begin moving at machine speed.Read All
The Joining Tax: A Year of Governing Claude Code
Multi-agent coding setups lose context at every handoff not because the design is broken, but because the boundary is the whole reason those roles exist and the handoff is its price. To make the seam cheap, you need capped review rounds, a priority order per role, structured handoffs, one owner for the issue list, and a doc-updater that closes the loop.Read All
DHH's Vibe Coding Advice Ignores the Biggest Problem With AI-Generated Code
DHH advocates vibe coding without review, but is there enough evidence to trust AI-generated software over the long term?
Showing 20 of 200 podcasts
Know a source we’re missing? Tell us about it