Agents should learn repeated work, but not by silently rewriting future runs.
Skill Workshop turns reusable agent lessons into reviewable proposals you can tweak, apply, or reject before they become live skills. https://t.co/g6TfHBi5NC
Hey @GeminiApp , please loosen the guidelines on video creation with avatars. I’ve tried to make some dad joke videos and I get blocked I guess because it’s too realistic?
Here’s my latest offensive prompt 🤷🏼♂️🤣
Create a highly realistic, cinematic local TV news segment in vertical 9:16 format. No captions, text overlays, logos, watermarks, or subtitles.
Main Character: (avatar)
He wears his usual outfit: black hoodie, brown pants, and black sneakers with white trim. He is acting as a professional television field reporter.
Setting
Outside a modern police station during daylight. Several marked police cruisers are visible behind him in the station parking lot. The cruisers are sitting up on blocks with all four wheels missing, clearly indicating the tires and rims have been stolen. Police officers move around in the background investigating the scene. The atmosphere feels like a legitimate local news broadcast.
0-2 Seconds
Wide establishing shot of the police station and the disabled police vehicles. Serious TV news music sting fades out.
The camera pushes in toward (avatar) holding a handheld news microphone.
2-8 Seconds
Medium shot.
(Avatar) maintains a completely professional reporter demeanor and looks directly into the camera.
In a serious local-news reporter voice, he says:
“A number of police cars have had their wheels stolen, and police are working tirelessly to catch the perpetrators. More at eleven.”
8-10 Seconds
The camera slowly zooms out.
Behind him, one officer notices another cruiser missing its wheels and throws his hands up in frustration.
(Avatar) remains completely serious and professional as the shot freezes for a beat like the end of a local news report.
@joshwoodward and team, please help lol I’ve also provided this feedback in app.
Workflows are the biggest upgrade to Claude Code’s capabilities since skills and subagents.
I dove deep into it with @sidbid to figure out best practices, examples and more. I’m particularly excited about the non-technical tasks it enables for Claude Code.
We’re making Codex more useful for your work by expanding plugins beyond individual tools.
These plugins turn Codex into a specialist for a specific role with a single install, no coding required.
Codex can access 62 popular apps and 110 skills for work across sales, data analytics, creative production, product design, and public equity investing.
https://t.co/nunrYP2uMI
Building apps has never been easier.
With Sites, Codex can turn your work, ideas, and plans into an interactive website or app your team can explore, use, and share with a URL.
Rolling out to Business and Enterprise plans, before expanding more broadly.
One of the biggest challenges companies are running into with AI right now is that individual contributors are often getting more leverage from AI than the organizations they work for.
Why?
Because the people doing the work know how the work actually gets done.
Most organizations have a procedure manual, a playbook, or a set of SOPs that describe how work is supposed to happen.
But let's be honest...
There's usually a gap between what the manual says and what actually happens on Tuesday afternoon when deadlines are looming, customers are waiting, and people are figuring things out in the real world.
I've worked with enough organizations to know that very few (if any?) have a perfect 100% match between documented processes and reality.
That's where AI gets interesting.
The people closest to the work are building prompts, workflows, automations, and AI agents around what actually happens. Meanwhile, many organizations are trying to deploy AI around what they think happens.
That disconnect matters.
It reminds me of an old manufacturing story.
A toothpaste company discovered some empty boxes were making it to customers. Leadership hired expensive consultants who designed a elaborate system that cost millions of dollars to implement, involving scales, alarms, and manual intervention. Empty boxes would trigger an alarm, an employee would remove the box, reset the system, and production would continue.
Problem solved.
For two days.
On the third day, leadership noticed there were no alarms at all. Curious, they went to investigate.
One of the employees responsible for responding to the alarms had placed a $10 box fan next to the conveyor belt.
The fan blew the empty boxes off the line before they ever reached the scale.
No alarms.
No empty boxes shipped.
Problem solved.
The lesson?
Sometimes the people closest to the work understand the problem better than the people designing the solution.
As companies rush to implement AI, they should spend less time asking, "How do we automate our documented process?" and more time asking, "How does this actually get done?"
Because AI built around reality will outperform AI built around assumptions every single time.
👊🏻
@emollick This is part of the problem. Individual contributions are automating the stuff in the right binder while companies are trying to automate the left binder across the organization. Until leaders acknowledge the delta and pivot accordingly, they’re gonna have a bad time.
Commentary is one of the most important pillars of X. And sometimes the best way to share your thoughts is with video.
Today we're launching a whole new way to make them:
React with Video
Tap the repost button and start recording with green screen, split screen, or picture-in-picture.
Now available on iOS
These two binders explain a lot about what’s happening with AI inside organizations right now.
If you’ve spent any time in a corporate environment, you know exactly what I’m talking about.
There is often a gap between documented process and actual process.
The manual says one thing.
Reality says something else.
Here’s why that matters in the age of AI.
Many organizations are rushing to deploy AI against the documented workflow. They’re feeding the policy manual into the AI, training assistants on official procedures, and optimizing processes that look great on paper.
The problem?
The paper version isn’t always how the work gets done.
The organizations pulling ahead are the ones talking to the people actually doing the work.
They’re mapping the real workflows.
The shortcuts.
The workarounds.
The handoffs.
The exceptions.
The tribal knowledge.
The stuff that never made it into the manual.
To be clear, I’m not advocating for violating policies, compliance requirements, or governance standards.
I’m talking about understanding reality.
Because AI doesn’t create leverage from fantasy.
It creates leverage from truth.
If your AI is trained on what the company says happens, while your competitors are training AI on what actually happens, guess who’s going to move faster?
The biggest AI advantage isn’t the model.
It’s understanding the real work.
The companies that close the gap between Binder #1 and Binder #2 are going to have a significant advantage over the ones pretending the gap doesn’t exist.
AI feels a lot like GPS did when it first became mainstream.
Most of the time it gets you where you're going faster. Every now and then it confidently tells you to turn somewhere you absolutely shouldn't.
Maybe you know the local high school lets out in 10 minutes. Maybe there's road construction that hasn't hit the data yet. Maybe you've driven that route a hundred times.
You don't ignore the GPS. You use it, then apply your own judgment.
That's where we are with AI.
The people struggling to automate everything aren't usually running into a technology problem. They're running into an experience problem.
If you don't know enough about the topic, you don't know when the AI is wrong. You don't know when to push back, redirect, or course-correct.
AI is incredibly powerful.
But the human is still the driver.
The loudest take online is that Hollywood hates AI.
I just spent a day with the filmmakers, execs, and studios actually using it - and the mood was the opposite.
We've hit a tipping point for AI getting used in real projects.
My takeaways on why the industry is shifting:
Suddenly on iOS 26.5 on an iPhone 16 Pro, my notifications that haven’t been cleared are batch sending to my Apple Watch (watch is 26.5) hours later. Blink camera notifications are biggest offenders. It’s pretty annoying. Anyone else experiencing this? @Apple