O'Reilly Media

08/21/2026

“Solving problems went cheap while selecting them became scarce.”

As AI agents get better at producing answers, the skills that set people apart are shifting: finding the right problems, knowing when the machine got it wrong, and taking responsibility for what ultimately ships.

In the agent era, judgment, taste, and the ability to finish strong may be some of the most valuable career skills we can build.

Check out the latest article by Addy Osmani to learn what it takes to build a durable career in the age of AI agents: https://bit.ly/4xfuCkD

AI gets good at anything with an answer key. Your career is everything that doesn’t have one.

08/21/2026

Google's Android earthquake alert system warned 11.4 million people in Venezuela before the ground started shaking, using nothing more than the accelerometers already built into their phones. Christina Stathopoulos breaks down how AI is doing real good in the world, from buying people precious seconds before an earthquake to mapping fast-moving wildfire boundaries in near real time through Google Maps and Search.

08/21/2026

Why do open models matter to you?

For Drew Breunig, CEO and cofounder of cmpnd.ai, that question connects to a problem many AI teams eventually run into: prompt debt.

What starts as a simple prompt can quickly become layers of instructions, edge cases, and contradictions, eventually leaving teams with systems that are difficult to change and locked to a specific model.

At AI Codecon, Drew’s session, “Avoiding Prompt Debt: Let the Model Write the Prompts,” will explore how to avoid that trap, use models to improve prompts, and put evals and tools like GEPA and DSPy to work.

Join us at AI Codecon to learn more.

Register for free: https://www.oreilly.com/AI-Codecon/

08/21/2026

“AI agents keep getting smarter, but the bigger story this week is how much they’re reshaping the systems around them. Models are optimizing less for chat and more for autonomous work.”

What happens when the primary user of AI models, and the web itself, stops being a person?

This week’s explores that shift, from models increasingly built for autonomous work to rising inference costs and agentic bots generating a growing share of web traffic.

As agents get faster, more capable, and better at coordinating, they’re not just changing how we use AI, they’re reshaping the economics, infrastructure, and security of the web itself.

Check out the article by Michelle Smith to learn more:

New frontier releases, an inference-first economy, and what happens when autonomous agents start coordinating with each other

08/21/2026

Knowing architecture concepts is one thing. Making a decision when every option has consequences takes practice.

The O’Reilly software architect cohort is built around that practice.

Over six weeks, you'll work through real-world architecture scenarios, make decisions, discuss trade-offs, and get feedback from Neal Ford and Mark Richards.

Only 25 people can take part, giving participants more opportunity to ask questions, contribute, and learn directly from the experts and each other.

This new learning format is available only on the O’Reilly learning platform.

Secure your spot: https://bit.ly/4ym4WDQ

08/20/2026

Anjuan Simmons has shipped software on physical disks, navigated the dot-com era, and watched cloud computing reshape the industry. He says the most durable thing those decades gave him is the judgment to sense "code smells," "organization smells," and "team smells," that tell him something's off before it’s put into words. That pattern recognition, built from watching things go wrong enough times to recognize the signs early, is what he argues no AI will replace in a tech leader.

08/20/2026

“Code quality is the symptom, not the disease. The deeper problem is epistemic agency: knowing what your system is doing and why. Lose that, and you lose the ability to make architectural decisions at all. You become a passenger in a system you built.”

As AI-generated code accelerates software development, a new challenge is emerging: cognitive debt, the growing gap between a system’s complexity and a team’s understanding of it.

In this article, Shreshta Shyamsundar explores how engineering leaders can prevent that gap from becoming “principal drift,” using risk-based task routing, independent review, comprehension checkpoints, and other practices designed to keep humans capable of steering the systems they build.

Check out the article to learn how teams can balance AI-driven velocity with understanding and control:

Navigating cognitive debt in AI-driven SDLC

08/19/2026

What happens when your buyer isn’t human?

“When software serves as the initial gatekeeper or negotiator, traditional relationship-building strategies yield diminishing returns during the agent’s screening process. The agent evaluates what it can measure: price, contract terms, delivery specifications, and compliance.”

AI agents are already changing how enterprise buyers research vendors, build shortlists, negotiate terms, and evaluate renewals. That shift has major implications for B2B sellers, and for commercial models designed around human buyers.

In this article, Rudrendu Paul and Sourav Nandy explore what an agent-first go-to-market model could look like, from outcome-based pricing and machine-readable product information to agent-compatible authorization.

Check out the article to learn more:

Enterprise sellers that redesign their pricing, sales motion, and customer success for agent buyers will now hold a structural advantage that slower competitors will never overcome.

08/19/2026

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—about AI, governance, data architecture, or whatever your team needs to learn next.

Comment below if we'll see you there.

08/19/2026

AI excels at pattern-driven work like writing code, but Anjuan S. argues that training a system on human judgment is a fundamentally different problem. What works for one engineer rarely works for another, and showing up for someone who's hurting or going through a crisis is too context-specific for any automated system to replicate. That’s why he suggests to learning the technical fundamentals and then focusing your energy on the things only a person can do.