OpenAI is not just racing on models anymore. According to WSJ, Greg Brockman is becoming more visible as a product-power operator inside OpenAI, as the company pushes ChatGPT, Codex, API work, and workflow agents toward the same agentic future.

OpenAI is reorganizing product power around the agent stack. As ChatGPT, Codex, the API, and AI workflow agents start to converge, the key question is no longer only which model is smarter. It is who can turn all those pieces into one operating surface for users and companies.

Microsoft says Copilot has standardized on Durable Task Scheduler to run long-running, multi-step AI workflows with state persistence, failure recovery, replay, and hundreds of millions of weekly executions.

This is more an AI infrastructure story than an assistant story. Once agents work across multiple steps, services, and failure points, the advantage is not just the prompt or the model call. It is orchestration: whether the system remembers state, retries correctly, resumes work, and stays controllable when AI work lasts longer than one chat turn.

TechCrunch, citing a memo viewed by The Information, says Meta is reportedly developing an AI-powered pendant and could begin testing it next year. The report builds on Meta's Limitless acquisition and its broader AI wearables push.

If true, this is another step in the AI assistant race: from chat apps to devices that sit much closer to the user. A pendant is not just a new form factor. It raises the question of who controls continuous personal context, and whether users will trade more privacy for an assistant that can remember, listen, and understand their environment better.

Anthropic dropped a zero-trust playbook for AI agents: Vaishnavi frames Anthropic's AI-agent security guidance as a zero-trust architecture playbook, not theory. The signal: once agents have tool access and run inside real workflows, prompt injection and access-control gaps become security architecture problems, not small prompt bugs.

Marc Andreessen's custom AI prompt becomes a prompt-quality caution signal: Marc Andreessen shared a long custom prompt asking AI to behave like a world-class expert and avoid hallucination. The signal is not whether that prompt is good or bad. It is that even power users are still trying to use instruction style to stabilize model behavior, while the missing layer is often verification, constraints, and workflow design.

AI job displacement is now explicit owner logic: Jason Calacanis says AI job displacement will be significant, and that recent layoffs are increasingly justified by owners who believe they can do more with fewer people. That points to a labor debate moving from "AI washing" toward real operating-model redesign.

Enterprises are reinvesting AI efficiency gains into new work: Aaron Levie offers the counter-signal: in many conversations with CIOs, CTOs, and CEOs, he sees companies growing because of AI or reinvesting efficiency savings into new functions such as FDEs and engineering. That makes the labor story less one-dimensional: AI may cut some work, but it can also redirect budget into new execution capacity.

AI fails when company knowledge stays trapped in people's heads: Tom Blomfield says using AI inside a company can feel like replacing employees with brilliant outsiders who do not know how the company works. The operator signal is sharp: model intelligence is not enough if domain knowledge stays inside people's heads instead of being packaged into context, rules, examples, and operating memory for AI.

CEOs are bargain hunting for cheaper AI: Axios says companies are watching AI bills more closely, routing tasks to cheaper or open models, and avoiding single-vendor lock-in. This is the cost-control phase of AI adoption: enterprise users are not only asking which model is best, but which tasks need expensive intelligence and which only need good-enough intelligence.

OpenAI publishes a playbook for trustworthy third-party frontier-model evaluations: OpenAI argues that frontier-model evaluations need to document the harness, tool setup, budget, validity checks, reward hacking, refusals, contamination, broken tasks, and sandbagging. For builders, the signal is clear: agent benchmarks are no longer simple scores. They are evaluations of the whole system around the model.

If you are choosing an AI agent for real work, do not start with the leaderboard. Start by asking what environment the agent was tested in. In 30 minutes, you can build a small scorecard that tells you whether an agent is ready for an actual workflow or just looks good in a benchmark.

DO THIS:

→ IF the agent only wins with broad tools, high budget, or many retries, do not call it a better model. Call it a context-heavy workflow.

→ IF it fails in the same way repeatedly, fix the harness before switching models.

→ IF the success metric is vague, every benchmark will become vibes.

→ IF the task touches data, money, customers, or production systems, keep the human approval boundary explicit before you automate.

Investors face a dilemma. When the S&P 500 finished its worst quarter since 2022 last month, diversifiers like bonds and bitcoin fell too.

Even with the turnaround in mid-April, analysts at Goldman Sachs and Vanguard have projected low-single-digit annualized returns from 2024-2034.

Bloomberg asked where experts would personally invest $100,000 for their March monthly edition.

One answer that surfaced for a second time? Art.

It's what billionaires like Bezos and the Rockefellers have privately used to diversify for decades.

Why?

Thanks to the world's premier art investing platform, now anyone can invest in works featuring legends like Banksy, Basquiat, and Picasso, without needing millions.

Shares in new offerings can sell quickly but...

Our subscribers skip their waitlist*

^{*According to Masterworks data. Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.}

WIRED looks at the agent wave through Claude Code, OpenClaw, builder enthusiasm, data and security risks, and the sense that agentic computing is becoming a new operating mode. Read it because it puts productivity, risk, usage behavior, and labor anxiety into one frame instead of simply saying agents are getting stronger.

VentureBeat argues that enterprise agents are entering a rebuild era: they need orchestration, observability, governance, recovery, and cost control. Read it if you want to understand why production agents cannot be just a model call plus a good prompt.

WIRED connects OpenAI's Codex push, Claude Code pressure, enterprise packaging, and the product reorganization around an agentic future. Read it to place the Greg Brockman/OpenAI product-power story inside the broader platform race.

TechRadar previews Microsoft Build 2026 as an event shaped around AI agents, developer frameworks, cloud and data, model training, responsible AI, and Windows. Treat it as a watchlist for where developer platforms may push next: agent infrastructure, local models, and responsible deployment.