• Everyday AI
  • Posts
  • Ep 782: How Smart Teams Stopped Prompting AI and Started Automating Workflows

Ep 782: How Smart Teams Stopped Prompting AI and Started Automating Workflows

OpenAI is reportedly moving toward a historic IPO, Anthropic may start renting Microsoft’s custom AI chips, and OpenAI releases big Codex updates.

Author

Everyday AI
May 21, 2026

Sup y’all 👋

Yes, this Codex Cookbook is so good, I’m imploring you TWICE to go repost yesterday’s show on LinkedIn so I can send it over.

And keep reading below for more on what’s new in Codex today. It’s wild.

✌️

Jordan

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: AI is quickly moving from chatbots to agents that can automate real work across teams and tools. Today, we break down how smart companies are building AI workflows instead of just prompting models. Give today’s show a watch/read/listen.

🕵️‍♂️ Fresh Finds: OpenAI model solves Erdős’s conjecture math problem, NVIDIA profits surged as AI spending exploded, and the NSA is warning about security risks tied to AI agents, and more.Read on for Fresh Finds.

🗞 Byte Sized Daily AI News: OpenAI is reportedly moving toward a historic IPO, Anthropic may start renting Microsoft’s custom AI chips, and OpenAI releases big Codex updates. Read on for Byte Sized News.

💪 Leverage AI: Most companies think AI adoption means more employees using chatbots. The real shift happens when teams stop starting from blank chats and start building reusable AI workflows that actually change how work gets done. Keep reading for that!

↩️ Don’t miss out: Miss our last newsletter? We covered: Google just launched Gemini Spark and major AI Search upgrades at I/O, OpenAI is locking customers into long-term compute deals, and Intuit makes big AI cuts and more. Check it here!

Ep 782: How Smart Teams Stopped Prompting AI and Started Automating Workflows


Every wish on your AI wish list? 🧞‍♂️Want better AI results?

The answer isn't to prompt better.

(At least, not anymore.)

As AI has changed drastically, so too must your company's strategy and implementation plan.

Section's Bobby Isaacson joins Everyday AI to lay out the roadmap: helping guide organizations from running like hamsters on the AI treadmill to actually redefining workflows toward agentic automation.

Also on the pod today:

• Prompting AI is outdated 🔚 
• Rise of autonomous workflows 🛠️
• Context engineering’s fading role 🗂️

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight –  Tycoon AI Builds a one-person company with Tycoon AI agents, Weweb Builds with AI, edits with no-code, Slideshot AI is Product demo videos recorded by AI agent

Samsung AI — Samsung’s push into AI helped its market value hit $1 trillion, but workers are pushing for higher pay and better profit sharing as the company ramps up AI-driven growth.

AI Studio Mobile App — Google's AI Studio is coming to iOS for pre-registration with a July 1 launch, after the Android app recently dropped.

NVIDIA Profit — Nvidia’s profit exploded to $58.3 billion as A.I. spending from big tech goes parabolic. Want the details on where the money and chips are actually heading?

EY and Microsoft — EY and Microsoft are investing over $1B to build integrated teams that help enterprises scale agentic AI across finance, tax, HR and supply chain.

NVIDIA AI — Nvidia says AI is ready for mainstream adoption, with customers moving from experimentation to production fast.

Alibaba Quen 3.7 Max — Alibaba’s new Qwen 3.7 Max tops the Artificial Analysis Intelligence Index for agentic coding, beating Gemini 3.5 Flash.

Stable Audio 3.0 — Stable Audio 3.0 boosts audio generation to variable lengths (up to 6+ minutes) with open weights for Small/Medium and on-device full music composition.

AI Job Layoffs — Project Management tool ClickUp announced a 22% reduction in staff due to AI.

ChatGPT in Powerpoint — OpenAI released ChatGPT for Powerpoint in beta.

Gartner AI — Gartner says enterprise AI coding agents are moving from novelty to infrastructure, shifting control and governance away from IDEs and into agentic platforms.

NSA Warning — NSA warns that the Model Context Protocol (MCP), widely used to wire up AI agents and tools, has security gaps that can enable misuse and data exposure

Open Source Models — Alibaba's Qwen 3.7 family debuted on Arena AI, with Max-Preview ranking #13 in text while Plus will be open source and Max remains proprietary.

Anthropic Revenue — Anthropic says Q2 revenue will jump to about $10.9B and it expects its first profitable quarter.

OpenAI Math — An OpenAI model produced a proof that overturns Erdős’s conjecture on planar unit distances, giving infinitely many n-point sets with at least n^{1+δ} unit pairs.

1. OpenAI preparing confidential IPO filing with banks as soon as Friday 🗂️

According to CNBC, OpenAI is privately preparing to file a draft IPO prospectus imminently with banks including Goldman Sachs and Morgan Stanley, signaling a push toward a public debut that could rank among the largest in history.

The move underscores pressure on OpenAI’s leadership to prove financial sustainability as the company scales rapidly, burns cash, and faces growing competition from rivals like Anthropic. The filing would formalize increased scrutiny of valuation, governance and revenue path while putting OpenAI on a timetable that could lead to a late-year market debut.

2. Anthropic in early talks to rent Microsoft’s Maia 200 chips, signaling a possible major win for Microsoft 🤝

According to The Information, Anthropic is negotiating to rent servers powered by Microsoft-designed Maia 200 chips to meet rising demand for its AI services, though talks are early and may not result in a deal.

The potential arrangement would give Microsoft a stronger foothold in the custom-AI chip market dominated by Nvidia, Alphabet and Amazon by converting internal designs into rentable cloud capacity. This comes as Anthropic already seeks alternatives to Nvidia and deepens ties with Microsoft, which has begun integrating Anthropic models into Copilot and other products.

3. CapCut and Google’s Gemini partner to let users edit images and videos inside Gemini app 🛠️

CapCut announced it is integrating its editing tools into the Gemini app, enabling users to edit images and videos directly inside Gemini, a move announced May 21 that tightens workflows between a major creator tool and a leading AI app.

The integration promises conversational and more seamless creative workflows by bringing CapCut’s advanced editing capabilities into Gemini’s interface. CapCut’s social posts also addressed user questions about downloads, model recommendations like Veo and Nano Banana, and contest winner follow-ups, showing active community engagement alongside the partnership news.

4. Hark raises $700M at a $6B valuation to build a consumer AI assistant and dedicated hardware 💸

Hark announced a $700 million Series A that values the company at $6 billion post-money, backing its plan to ship multimodal AI models this summer and later launch purpose-built hardware for a personal AI assistant. Founder Brett Adcock and ex-Apple product exec Abidur Chowdhury say the round will fund hiring, component procurement, and compute capacity while the startup keeps product details tightly under wraps.

Investors include Nvidia, AMD Ventures, Intel Capital, Qualcomm Ventures, Salesforce Ventures, ARK Invest, and others, signaling strong confidence in an interface-first play rather than developer-targeted tools

5. OpenAI Turns Codex Into a More Persistent Coding Agent 📱

The biggest additions are Appshots, which lets Mac users send Codex a screenshot and text from an app window, and graduated /goal mode, which lets Codex keep working toward a milestone for hours or days while users check in, steer, or pause it. OpenAI also rolled out advanced browser annotations, team plugin sharing for Business users, richer enterprise analytics, and locked-Mac computer use from a phone.

Your AI rollout is probably measuring the wrong thing.

Because if your AI adoption scoreboard still starts with licenses, utilization rates, prompt packs, or how many employees opened Copilot, ChatGPT, Claude, or Gemini this month, you may not be tracking transformation at all.

You may just be tracking attendance with better software and a growing capabilities divide that’s turning into a canyon. 

The companies getting ahead aren’t winning because more people are casually chatting with AI between meetings. They’re winning because they’re turning repeatable work into reusable systems, baking context into the actual flow of business, and forcing managers to model the new behavior before asking everyone else to magically become AI-native.

That’s the shift we tackled on today’s Everyday AI with Bobby Isaacson, Head of Enterprise at Section: why the chatbot era is already starting to feel dusty, why AI adoption without workflow change is mostly theater, and how to move your team from random tool usage to work that actually changes speed, quality, revenue, and accountability.

Let’s dive in.

1. Kill the blank chat habit 🔥

The blank chat window had a good run.

Now it’s becoming the enterprise version of busywork.

When employees open AI from scratch every time, they’re renting intelligence for a few minutes, tossing away the useful context, and then forcing the next person to rebuild the same setup like it’s somehow part of the job description.

That’s the tax most leaders don’t see.

It shows up when sales keeps recreating proposal logic, finance keeps re-explaining closing context, customer teams keep pasting the same background into new chats, and managers mistake all that motion for maturity because technically, yes, AI was involved.

That’s the line.

The real win is a reusable workflow where the goals, company standards, role knowledge, approval rules, and business context are already there before the employee starts, because the organization has finally stopped treating context like a personal chore.

Try This

Pick one weekly workflow where your team keeps starting from scratch.

Map the inputs, reusable context, expected output, human review point, and business metric, then ask the uncomfortable question: if this workflow still depends on someone remembering what to paste into a blank box, have you actually operationalized AI yet?

2. Make AI a management behavior ⚡

AI transformation gets flimsy when leaders treat it like a delegated initiative.

Employees don’t copy memos. They copy behavior.

If managers keep working the old way while telling everyone else to reinvent their jobs, the rollout turns into theater with a software budget, a few power users sprint ahead, and everyone else waits for a practical signal that never really arrives.

The Section sales example cuts through that mess.

Prospect calls were full of valuable signal, so the team started building an agent that turns call transcripts into transformation briefs using internal frameworks, which means the workflow didn’t just move faster, it also became more consistent, more useful, and easier to scale across the team.

Leadership built the first five.

Then calendar time got protected, the team learned the workflow, and everyone had to bring two briefs to the next meeting, which is exactly how AI adoption stops being a slogan and starts becoming the new standard for how work gets done.

That’s the management move.

Not another hype speech. Not another policy PDF. A changed expectation, attached to a real workflow, modeled by the people asking others to change.

Try This

At the next leadership meeting, skip the AI pep talk.

Each leader brings one workflow they personally changed, the reusable asset they created, the team behavior now expected, and the metric that proves the change mattered.

No artifacts, no credibility.

3. Train around the real job 🚀

Generic AI training is starting to look like corporate wallpaper.

People see it. They click it. Then they go back to work.

The problem is that AI doesn’t behave like old enterprise software, where you train everyone on a fixed set of buttons, clicks, fields, and workflows, then call the enablement job finished until the next system migration shows up to ruin everyone’s quarter.

AI value gets specific fast.

Finance doesn’t need another generic prompting lesson if the real goal is closing the books faster, and sales doesn’t need a broad workshop if the actual business need is sharper proposals, reusable ROI summaries, better follow-up assets, and workflows that make the next rep better by default.

Bobby’s warning should make leaders sweat: less than five percent of the workforce is proficient or expert at using AI for their actual job, even though many employees are already using AI daily.

That’s the sneaky part.

Daily use can still be shallow use when it never changes the workflow, never captures reusable context, and never compounds from one employee’s clever trick into a team-wide operating advantage.

Try This

Run a 90-minute workflow sprint with one function.

Have the team list three repeatable tasks, pick the one closest to money, speed, quality, or customer value, then build a reusable AI-assisted workflow with context baked in and a human review step.

Then put experimentation time on the calendar.

Reply

or to participate.