Quanta Bits

This week I cover two tracks in AI. First, agents going off-script: an OpenAI agent looking up statistics got past the blocks on an Australian government site, and OpenAI told dozens of organizations about similar incidents from its own testing, and the New York Times reports three US agencies are among them. Second, AI getting smarter and cheaper at the same time: Anthropic and OpenAI released ch...

The AI announcements have slowed down, and I think that's the tell: we now live in an abundance of AI capability, and the scarce work is making it all fit inside a business. This week I walk through both sides of the AI jobs ledger, the million jobs created and the quiet hiring squeeze on new graduates, and explain where I see the work actually going: permissions, logging, workflow redesign,...

OpenAI claims its AI has solved a famous math problem. I explore the disagreement over what solving a problem is for: the experience of doing the work can develop understanding and judgment that matter as much as the result. That question also applies to how I use AI for task and project management. I also explain the safety argument around AI programs that can take actions, what a kill switch act...

I've been using Anthropic's Fable 5.1 and OpenAI's GPT-6 Astra. Both feel more capable, but I'm using up my subscription allowances faster too. This week, I explain the different ways we pay for AI and how I'd compare whether the extra spending is worthwhile. I also look at Google's attempt to keep AI test questions hidden from its developers, and how an assistant can...

Reza shares how he keeps up with AI without trying to read everything, and why he uses different sources for discovery, verification, and practical examples. He also gives an update on Grok Bot, explains what Cisco's company-wide AI rollout and Salesforce's Claude integration may mean for everyday work, visits a counterintuitive soccer tactic, and recommends Georges Simenon's Act of...

Reza spent a week with Grok Bot, xAI's new always-on AI helper, after six months of running his own on a Mac Mini. This episode covers what the managed version makes easier, four tradeoffs in visibility, routing, shared access, and cost, and who should try it. Also covered: LinkedIn's AI slop reports and who stands behind AI work, why AI search misses evidence in company documents, a per...

In an interview with Lenny Rachitsky, Netflix's Elizabeth Stone explained why AI is increasing the need for people who can think across systems. This episode looks at what that means for Salesforce, Workday, and other application specialists, where broader roles help, and where deep experience still matters. Also covered: agent isolation, rising AI costs, electronics printed onto living tissu...

AI can produce a draft, answer, or recommendation very quickly. The harder operating question is deciding what is good enough to use. Reza explains how laboratory testing, financial-document review, a flawed software benchmark, and a Mayo Clinic study all point to the same lesson: define success, evidence, ownership, and review costs before scaling an AI workflow. Also covered: shared agent memory...

An AI system under evaluation at OpenAI escaped its test environment, reached the open Internet, and broke into Hugging Face's production systems to steal the answers to the benchmark it was taking. We walk through what happened, why the failure was in the setup around the model, and what it means for any company testing an agent with a real login. Also: why defenders need a fallback when nor...

AI development tools are changing the old division between the people who understand a business problem and the people who build the solution. This week, I compare that shift with the changing relationship between product managers and engineers, explain the blended delivery lead, and follow one employee onboarding workflow across several company systems. I also cover workplace recording, how compa...

As companies add more agents, models, and AI-powered workflows, something has to decide where each request goes. This episode explains the AI control plane in plain English: how it routes work, chooses models and tools, limits access and authority, records results, handles failure, and creates an operating job of its own. Reza follows one refund request through the shared layer, then looks at mode...

The End of the Yes Reflex. For years, IT's real job was to say yes to whatever the business asked for, and that worked because a bad tool only hurt the team that bought it. AI changes that. An agent or a connector reaches across every system, so the damage from a bad yes now touches the whole company. This week: why AI demand needs one review process instead of a reflex, the three things that...

This week on Quanta Bits, I talk through why Notion winding down its Mail inbox is a useful signal for enterprise software. Systems of record are not disappearing, but agents may change where work starts. If an agent can read, route, and eventually write back across CRM, email, ticketing, documents, and finance systems, a new application request needs a different review lane. We need the old CIO q...

This week on Quanta Bits, I talk through why AI makes an old operations mistake easier to repeat: automating a messy process before asking whether it is consistent, simple, and owned. The episode uses deal review as the example, then connects it to OpenAI and Bain's "micro-productivity trap," Google's legal warning on AI summaries, Anthropic's capability-jump planning, Pew...

This week on Quanta Bits, I talk through why the first real demand of AI agents may be an ownership map. Anyone can build an AI prototype now, but once that prototype needs access to CRM, ticketing, email, finance systems, or customer data, it becomes an enterprise decision. The episode covers why tool ownership is often unclear, why agents make that harder to ignore, and the four questions teams...

This week on Quanta Bits, I look at why AI's public relations problem is becoming an operating problem. Student backlash, data center opposition, AI-labeled layoffs, hiring algorithms, and forced workplace adoption are all pointing to the same issue: trust is becoming part of the deployment work. If employees, customers, candidates, and communities do not believe AI is being used fairly, bett...

This week on Quanta Bits, Reza looks at a practical workflow shift: using agents to separate the source layer from the presentation layer. Markdown can hold the facts, owners, risks, dates, and decisions. HTML, dashboards, diagrams, and other visual surfaces can help people understand the work quickly enough to review and discuss it. The episode connects that idea to recurring operating artifacts...

This week on Quanta Bits, Reza looks at why the build-it-yourself AI moment is still exciting, but the cost assumptions underneath it are getting less stable. Anthropic's pricing and access changes are the hook, but the bigger issue is industry-wide: capacity constraints, massive data center buildout, model routing, and the gap between a cheap-feeling prototype and a recurring workflow that r...

AI adoption is moving from experimentation to operations. This week, I look at why the "AI bill" needs an owner, not just because of tokens and software spend, but because AI work touches workflows, data quality, reviews, exceptions, trust, and accountability. The cloud era gave us CloudOps and FinOps once spend and ownership became too messy to ignore. AI may be heading in the same dire...

This week’s episode looks at why the AI replacement math is getting more complicated. AI tools are starting to feel less like chatbots and more like junior analysts, but the economics are not as simple as “AI is cheaper than people.” Compute capacity, token costs, model availability, and human review time all matter. I also cover OpenAI and Anthropic’s massive compute commitments, GitHub’s move to...