Anthropic’s Barry Zhang recently offered the clearest taxonomy I have seen on what an agent actually is, what a workflow actually is, and when each one belongs in production. If you are wondering what your team should be building this quarter, this framing will be extremely helpful.

OpenAI just shipped GPT-5.5, seven weeks after GPT-5.4, and OpenAI President Greg Brockman called it “a new class of intelligence.”

OpenAI launched "workspace agents" in ChatGPT yesterday. Powered by Codex, these persistent, autonomous agents run in the cloud, integrate with ChatGPT and Slack, and handle complex workflows across tools and teams.

The industry is splitting into two camps. Which camp is right?

The buried story is what Apple told us about the next decade by picking Ternus: they picked an engineer. That is a strategic declaration, even if no one at Apple will say so out loud.

Every company deploying AI agents needs evaluation frameworks to grade output quality. When an agent drafts an email or generates a report, something/someone needs to decide if the result is usable. Without explicit evaluation criteria, you get technically correct outputs that feel (or are) wrong.

For devs, Opus 4.7 is a genuine step forward. For consumers, the story is a bit different.

Google just launched Skills in Chrome, which converts your AI prompts into reusable one-click tools. I like this approach because it acknowledges how people use AI tools. Essentially, you are programming Chrome to repeat your most useful AI interactions without rebuilding the context each time.

Microsoft is exploring ways to integrate OpenClaw-style autonomous features into its Copilot AI assistant, according to The Information. Omar Shahine, a corporate VP at Microsoft, confirmed the company is “exploring the potential of technologies like OpenClaw in an enterprise context.”

Every company has cultural debt. Few are malicious, but all of them are expensive… and they get dramatically more expensive when they are encoded into the DNA of your hybrid human/AI workforce. Read more at the link below.

Are managed agents the SaaS business that survives SaaSpocalypse?

Do we let private companies own foundation models that are powerful enough to take down critical infrastructure? Anthropic just made the case for nationalizing themselves by acting like a responsible lab. Will other AI research firms make the same choice?

If any of your enterprise workflows depend on third-party AI services that big tech can easily replicate, make sure your contracts are short-term.

Murphy Campbell, a folk singer from western North Carolina with 7,800 monthly Spotify listeners, discovered someone cloned her voice, uploaded fake songs to her streaming pages, and then used copyright claims to steal revenue from her original YouTube videos. Think about this...

Will free AI video generation become table stakes across the industry? This is exactly what the SaaS-pocalypse doomsayers predict.

If this approach gains traction, we’ll be optimizing for dialogue, not clicks, and planners and buyers will have to invent a way to negotiate conversational ad performance metrics that don’t exist yet.

For Anthropic, this incident is particularly awkward given their positioning as a safety-focused AI company. It’s not existential, but it’s another lesson on the road to AGI.

I’m not sure how legal it would be to robocall every bar in New York, but in Dublin the economic impact was immediate.

My essay explores the emerging world of "managed agents." When your sales agent queries the CRM and your finance agent hits the ERP system, someone has to define what each can see, touch, and change. Managed agents are about to become "the" leadership challenge of our time.

Love and respect to the dedicated human wiki-police. You’re doing more important work than you know.

Anthropic just shipped “auto mode” for Claude Code. According to the company, users approve 93% of permission prompts without reading them. The guardrails had become a rubber stamp, and everyone knew it.

OpenAI has not explained why it’s shuttering Sora or what (if anything) went wrong with the Disney partnership.

Start using computer use features slowly. It’s going to take a while to understand what agents/claws/personal AI assistants can and can’t do. More importantly, figure out what your agent(s) should and shouldn’t do.

After calling OpenClaw the most successful open-source project in history, Jensen Huang said that every enterprise company and every software company in the world needs an agentic strategy, and specifically needs an OpenClaw strategy. My Sunday essay, Corporate Claws, explains what he means.

At NVIDIA GTC, after calling OpenClaw the most successful open-source project in history, Jensen Huang said that every single enterprise company and every single software company in the world needs an agentic strategy, and specifically needs an OpenClaw strategy. shellypalmer.com/2026/03/corp...

Bot-to-bot, agent-to-agent, claw-to-claw… call it what you will. The data says we’re headed toward an internet where machines talk primarily to other machines and humans occasionally participate.

This feels like a temporary win rather than a permanent solution. The government says it will take “the time needed” to balance artists’ wishes with AI development. That careful phrasing suggests they’re looking for a middle path that keeps both sides happy.

I’ve been watching the agentic space evolve rapidly. The shift from chatbots to autonomous assistants requires new infrastructure; NVIDIA is positioning itself as the platform provider. This move extends their GPU dominance into the software layer where AI agents actually live and work. I love it!

If you’re relying on manual tracking or honor-system budgets, Google just gave you another reason to operationalize. Today would be a good day to ask your team about how they set per-project caps, configure usage thresholds, and build cost attribution into your AI reporting.

AI companies are now using AI to build AI products, and the acceleration is visible in weekly release cycles that used to take months. Anthropic built its entire Cowork productivity tool using Claude in just a week and a half. This is recursive self-improvement in practice.

It’s time to stop thinking in text and start thinking multi-modally!

Nvidia released Nemotron 3 Super on Wednesday, a 120-billion-parameter open model that activates only 12 billion parameters at inference time. The numbers matter, so let’s walk through the math.

Amazon just put Health AI on its main app. No Prime required. The key difference vs. competitors: it connects to your actual medical records and doctors who can prescribe. shellypalmer.com/2026/03/amaz...

Yann LeCun raised $1.03B to bet against the entire generative AI playbook. AMI Labs will build "world models" that understand causality as well as of correlation. Not sure what scares me more, AI that misunderstands how the world works or AI that understands it? shellypalmer.com/2026/03/yann...

Running a personal AI claw on Claude Opus costs $300-$500/mo. Scale that to 500 employees and you're at $250K/month before governance. The per-function model is where most enterprises will land first. Here's why. shellypalmer.com/2026/03/depl...

GPT-5.4 is OpenAI’s first general-purpose model with built-in computer use capabilities.

The math isn't mathing.

“Maybe you think the Iran strike was good and the Venezuela invasion was bad,” Altman said in an all-hands meeting. “You don’t get to weigh in on that.”

The Supreme Court will hear an AI copyright case eventually. It won’t be this one. Until then, the rule is what it’s always been: humans create, machines assist. Act accordingly.

My Sunday essay walks through what to ask your tech team this week and why most companies don't know which of their SaaS vendors are white-labeling Claude under the hood.

Some will call it AI Slop, others will call it AI Candy, and still others will recognize it for what it is: the next generation of imaging tools.

You can imagine a future Burger King run by AI lording over the smallest possible number of low-paid human workers. That’s not exactly having it your way.

If you haven’t gone deep into the Claude desktop app, you should give it a look. The three modes (Chat, Cowork, and Code) serve different purposes. Understanding what each one does will help you understand why this new plugin marketplace is an important next step in leveraging AI for the enterprise.

What is a 76.8% verbatim reproduction of Harry Potter?

I've continued to refine a concept I call "share of prompt." It's a three-layer measurement framework that connects AI answer engine visibility directly to revenue. Every major analytics platform shipped AI visibility tools in 2025. None can tell you what an AI-generated brand mention is worth.

You are fully entitled to feel however you want to feel about this, but don’t think for a second that these capabilities will be rejected by SMBs or by professional production organizations for all but the very lowest-end work. Quite conversely: the adventure is just beginning.

The near-term impact of Lyria 3 lands squarely on production music, which is a billion dollar market filled with tracks designed to be heard but not listened to.

We’ve been talking about marketing to bots for the past year or so. The infrastructure to optimize for agentic commerce is almost here. Get ready.

While OpenAI, Google, and Anthropic fight over who can build the biggest brain, Cohere went small and wide.

Rhett Reese’s world may be "over," but for everyone else with a story to tell and a laptop to tell it on, it is just beginning.