O'Reilly

We just released new UK research revealing that training and governance have failed to keep pace with AI-assisted work, leaving employees exposed to costly mistakes. AI has quietly become the first place most UK workers turn for answers: https://bit.ly/4zRIgLt

“The distilled model wasn’t just as good as the model+harness, it was noticeably better.” Harnesses are evolving as models absorb general scaffolding. They still handle what’s specific to your tools, data, users, and environment. Read the #Radar article by Laurie Voss: https://bit.ly/481XB0p

AI agents can make decisions quietly and autonomously, sometimes before anyone realizes what’s happening. AI Horror Stories explores the risks and what to do when things go wrong. Sign up today: https://bit.ly/4Bup0Wa

“These are modes, not personality types.” Sar Haidar explores five ways people use AI at work, and how teams can keep people and agents aligned. Read #Radar article to learn more: https://bit.ly/4yDeqJH

“A software factory lets you delegate more of that process!” Hugo Bowne-Anderson and Eleanor Berger on agents that work in the background, with context, automation, and checks. Read the #Radar article: https://bit.ly/4e0ZjT7

AI should help people create and flourish. @timoreilly.bsky.social and Demetrios Brinkmann explore human-centered AI, and how expert context changed an agent’s hiring advice. Watch the full conversation: https://bit.ly/4ij0fn3

“AI’s rapid expansion is creating problems across the industry, from chip shortages to autonomous agents that exceed their instructions.” Christina Stathopoulos and Michelle Smith cover the AI’s growing pains. Read the #Radar article here: https://bit.ly/4ss3DR5

“You don’t need a Ferrari to manage your calendar.” — Raffi Krikorian Mozilla CTO Raffi Krikorian and @timoreilly.bsky.social explore open source AI, and building technology people can own and trust. Read here to learn more: https://bit.ly/3ULecCs

Coming up on 10/28 from @oreilly.bsky.social & @techupskills.bsky.social, our popular online training "Implementing #AI #Agents in #Python" from noon - 5 pm EDT. Learn what agents are and how to build them, along with #MCP servers, true agentic #RAG & more. See www.oreilly.com/live-events/...

Big ole WHOA moment. A whoament if you will (this is @timoreilly.bsky.social, founder of @oreilly.bsky.social and OSS OG)

Your AI text can carry a watermark without adding a single hidden character. “There are no hidden characters placed inside the AI generated text, and no extra tokens are consumed.” Bruce Hopkins explains how it works in this #Radar article: https://bit.ly/3UHIAxx

Curious about generative AI, but not sure where to start, especially if your role isn’t technical? Take the Generative AI at Work assessment to get a skill plan tailored to what you need to learn. Get started here: https://bit.ly/4ywJvyM

“AI is also a tool for defenders.” But security teams need more than copilots: clear policies, specialized tools, funding, and new skills. Mike Loukides on the AI security gaps: https://bit.ly/47xFvmE

Feeling like your AI progress has stalled? Move beyond one-off wins and build repeatable workflows that help you execute better. Join O’Reilly’s Expert Showcase: Building AI Workflows That Stick on November 10. Register: https://bit.ly/47qyTXb

A first successful AI-assisted run is weak evidence. Can someone else reconstruct, question, or change the work without the creator’s private context? Kaiyao Duan on why the “second-user test” reveals real understanding: https://bit.ly/3W0mWF0

The 2026 O’Reilly Award winners are here! This year’s honorees show what it looks like to turn learning into action, with standout work spanning onboarding, skills transformation, applied AI, engineering, and more. Meet the winners: https://bit.ly/4z5dcIq

“Wider access to AI tools will make human expertise in interpreting data, evaluating models, and understanding their limitations increasingly valuable.” New tools, robotics, and AI research: this week’s #Radar roundup by Christina Stathopoulos and Michelle Smith. Learn more: https://bit.ly/47jSp7E

“Treat blast radius as a review criterion on its own.” AI-generated code may be fast, but every change still needs evidence, clear scope, and the right level of review. Check out this #Radar article by Markus Eisele to learn more: https://bit.ly/4dL7VNw

On 10/20, Alex Komoroske joins @timoreilly.bsky.social to explore resonant computing, an approach to technology built around helping people act in ways that reflect their deeper values. They will explore why platforms behave more like complex ecosystems than products: https://bit.ly/4chZk4H

“Observability is for unknown unknowns.” As AI accelerates software delivery, teams need better feedback loops to understand the unexpected. Charity Majors joins Ben Lorica to discuss reliable AI-powered systems. Check out the #Radar conversation to learn more: https://bit.ly/4sa1Os7

“Software quality now depends on the constraints you set around your agents.” As AI agents outpace traditional code review, quality gates matter more than ever. Check out this #Radar article by Addy Osmani to learn more: https://bit.ly/4hW90mH

“Once an LLM generates a word, it’s part of the response.” Why do LLMs lapse into “Yoda-speak”? Andrew Stellman explores how attention and autoregression shape the AI writing quirks we know too well. Read more: https://bit.ly/4jWNQYc

On Dec 4 I'm joining Vicki Reyzelman on @oreilly.bsky.social's Beyond the Book to talk AI Agents with MCP: the core architecture, building servers and clients in Python, and agent designs that hold up as models change. Live Q&A, 10am ET / 7am PT.

What makes an architectural decision useful to an AI coding agent? “Retrieval finds; applicability governs.” Theo Valmis on why relevant context isn’t necessarily binding, and why guidance and enforcement need different roles. Read the #Radar article: https://bit.ly/4hLiken

“OpenClaw supports both, and you decide where the agent runs, which models it uses, and how it accesses your data.” Read Michelle Smith’s #Radar article on building a personal assistant: https://bit.ly/4i2QCsG

Heading to San Jose next week for the @aaif.io AGNTCon + MCPCon. I'm signing AI Agents with MCP at the @oreilly.bsky.social booth during Thursday's AM coffee break (10:10-10:35). I'll have 20-25 copies on hand, and I'll sign any you bring too. Come say hi!

“WebMCP gives web developers a standard way to publish structured tools to access the capabilities of their website enabling agents to call those functions with structured inputs.” How can websites work better for people and AI agents? Learn more: https://bit.ly/4zxRgpc

"AI systems are gaining more autonomy while governments, companies, and researchers are still working out how much oversight they need." This week’s AI news underscores a key reality: more capability also means more responsibility. Learn more: https://bit.ly/4hCrk5w

Identity verification can tell you who an AI agent is but not whether its request is safe. “The signature proves who is making the request, not the intent behind it.” Vicki Reyzelman on why securing AI agents requires more than identity checks: https://bit.ly/4rSN8gE

Are your engineering metrics measuring real value or just activity? Join Rachel James, O’Reilly's Engineering Director, for LeadDev’s Oct. 22 panel: “The truth about engineering team performance in 2026.” Register today: https://bit.ly/4rXSGX

On 10/15, Andrew Odewahn will join @timoreilly.bsky.social to explore MCP, agent skills, stakeholder review, and how ongoing human-AI conversations can help organizations build and refine knowledge over time: https://bit.ly/4chZk4H

Join host Chester Ismay every other Monday throughout the 2026 football season for The Monday Morning Agent, where you’ll use real-life data to settle enduring football debates while gradually building an AI data analyst capable of doing that on its own: https://bit.ly/4yGtZ4n

AI moves fast. Can we explain a past decision? “Reconstructability is a system property established during execution, not inferred later from surviving artifacts.” Monika Dvorackova’s #Radar article: https://bit.ly/3VYPbUx

Learn more about our Expert Intelligence, an organization-wide suite of tools that grounds enterprise AI in O’Reilly’s deep repository of practitioner knowledge and calls on job-specific agent skills to generate actionable guidance for real-world work: https://bit.ly/3VCPAvI

How do you turn a base LLM into a useful assistant? “But no amount of tooling fixes bad data or a reward model that scores the wrong things highly.” Sharon Zhou walks through SFT, reward modeling, and PPO: https://bit.ly/3Vwwc3G

That’s a wrap on six weeks of Software Architecture! Participants explored patterns, distributed systems, evolutionary architecture, and more with Neal Ford and Mark Richards. Get updates on future cohorts: https://bit.ly/4ym4WDQ

AI agents are becoming cheaper, more specialized, and more autonomous but accountability can’t be delegated. “AI has infinite patience and is fundamentally probabilistic.” Read Mike Loukides’s October #Radar roundup: https://bit.ly/4hIx2BL

Over 20 sessions with instructor Kesha Williams, you’ll build one thing: an AI code reviewer that runs in your CI pipeline, reads real pull requests, and posts inline comments at the exact file and line. Register to attend: https://bit.ly/4zfEM54

“You don’t need to replace the systems that already handle your application logic to help them work with agents.” Bruce Hopkins shows how to add an MCP layer around a stock-data API. Read Michelle Smith’s #Radar article: https://bit.ly/3W7Kbgi

“Prompt tuning and stronger models can improve the agent’s responses, but can’t provide guarantees for actions.” Reliable agents need enforceable policies, narrowly scoped authority, and recoverable state. Read Angie Jones’s #Radar article: https://bit.ly/4j1doDe

Five weeks down, one to go! This week, our Software Architecture cohort dove into evolutionary architecture and Architecture as Code as they prepare for the final session. Interested in future opportunities? Add your name: https://bit.ly/4ym4WDQ

From my @oreilly.bsky.social Radar article, “The AI Teaching Toolkit: Practical Guidance for Teams” www.oreilly.com/radar/the-ai...

Coding agents need project context without a courtroom transcript. “Each governing ADR should describe the decision currently in force.” Read Duncan Davidson’s article on keeping decision records useful for agents and humans: https://bit.ly/4rGXPD7

“Deploying an agent means taking responsibility for the systems around it.” As agents move faster and act with more autonomy boundaries become critical design choices. Read Michelle Smith’s latest article: https://bit.ly/3Vt0tjF

On October 7th, I’m running a new #OReilly course: AI-Powered Excel Charts. Find out more and register: learning.oreilly.com/live-events/... @oreilly.bsky.social

Coming up in October from @techupskills.bsky.social , our popular @oreilly.bsky.social courses on #GitHub #Copilot & Implementing #AI #Agents in #Python. Also NFJS ones on Fine-Tuning Small AI Models, AI #Security & #Prompt Eng. Details in video & at techskillstransformations.com/training-cal... !

Coming up in October from @techupskills.bsky.social , our popular @oreilly.bsky.social courses on #GitHub #Copilot & Implementing #AI #Agents in #Python. Also NFJS ones on Fine-Tuning Small AI Models, AI #Security & #Prompt Eng. Details in video & at techskillstransformations.com/training-cal... !

“For decades, people have adapted to computers. It now seems possible that software may finally start adapting to people instead.” Check out this #Radar article by Robert Englander to learn more: https://bit.ly/4rEy85Q

We'll be at Gartner IT Symposium/Xpo in Orlando, Florida, October 19–22, talking with technology leaders about the skills their teams need to keep pace with AI. Add the O'Reilly booth to your agenda and bring your questions.

On October 5, Rohit Krishnan will join @timoreilly.bsky.social to share what has experiments reveal about human-agent coordination and how organizations should divide work between people and agents. Register here: bit.ly/4chZk4H