Last month, a senior brand strategy exec I know pulled me aside at an event because she didn’t know how to handle a very 2026 problem.
Someone on her team had sent her a strategy.
Twenty-three pages. Organized sections, beautifully formatted, professional language.
On the surface, it was the kind of deliverable that passes the first-glance test.
She started reading it, expecting to engage with the thinking, push back on a few points, and sharpen the recommendations.
But by page 2, she realized there was nothing to engage with.
The strategy was poorly thought through.
The “insights” were generic observations dressed up in confident language.
The reasoning didn’t hold up.
And the recommendations weren’t defensible.
So, she did what a lot of managers are doing right now.
She absorbed the cost and spent her evening rewriting it herself. For someone who reports to her.
When I asked what she said to the person, she paused.
“Nothing. I honestly don’t know what to do. I don’t want to tell them not to use AI. I want them to use AI. I just... want them to actually think.”
There’s now a word for this type of output: workslop—AI-generated content that looks like real work but doesn’t actually move anything forward.
According to a study from Stanford and from BetterUp, 40% of workers said they’ve received workslop from peers, direct reports, or even their managers. And each instance takes nearly two hours to address.
When asked about their experience with workslop, one project manager said:
“Receiving this poor quality work created a huge time waste and inconvenience for me. Since it was provided by my supervisor, I felt uncomfortable confronting her about its poor quality and requesting she redo it.
So instead, I had to take on effort to do something that should have been her responsibility, which got in the way of my other ongoing projects.”
At scale, this costs large companies millions a year.
But the cost to anyone sending workslop is even higher.
Half of recipients view colleagues who send workslop as less creative, capable and reliable.
42% see them as less trustworthy.
And no one sends you a note that says, “Just so you know, after reading that deck, I trust your judgment a lot less now.”
Your work gets scrutinized more carefully.
Your name comes up less for the projects that matter.
The shift is silent, and by the time you notice, it’s already reshaped how people see you.
Why is This Happening?
Most employees are now expected to use AI for their work in some capacity.
Many got minimal training. Some got none at all.
Organizations pushed the tools without defining what good AI-assisted work actually looks like, and people filled that gap with their best guess—which, when it comes to evaluating AI output, is often not enough.
It doesn’t help that the people assigning the work often don’t understand how AI tools work either, and their direction reflects that.
When a manager’s instruction is “just run it through ChatGPT,” that’s like handing your team a recipe for workslop, because AI outputs reflect the quality of what you feed it.
There’s also something genuinely tricky about AI that I don’t think enough people talk about.
When you generate something and it comes back looking polished, with clean grammar and professional structure, your brain registers it as done.
The output ‘looks’ finished.
And when you’re buried in deadlines and asked to use AI without proper training, it becomes easy to mistake polished for ready and hit send.
The result is mediocre work produced at a speed and volume that looks like productivity, but actually creates more problems than it solves.
Average is Easy Now
I’ve used AI every day for over three and a half years.
Claude and ChatGPT are a foundational part of how I research, analyze, pressure-test ideas, and think through problems.
And yes, they save me a lot of time on some tedious parts of the work.
But for many parts of my work, AI often slows me down before it makes it better.
Because now I can explore more angles. Test more assumptions. Spot the gaps. Ask better questions. And push harder on the logic.
And that takes time.
AI is very good at getting you to average quickly.
But when everyone can get to average quickly, average is basically worthless.
The value is in what only you can bring to it.
Your expertise. Your judgment. Your taste.
Your understanding of the client, brand, audience, politics, timing, tradeoffs, and all the messy context that never fits neatly into a prompt.
That’s your edge.
And AI amplifies every bit of it.
Bring your depth and learn to wield AI, and you’ll become incredibly hard to compete with.
Out-AI Them
Back to the twenty-three-page strategy.
I told the exec something that might sound counterintuitive: out-AI them.
She knows what a solid strategy looks like. She’s been developing and evaluating them for years.
She has clear criteria for what makes insights useful, recommendations defensible, and analysis rigorous.
But a lot of that knowledge lives in her head.
So the first step is making her standards explicit enough that both her team and the AI can check against it before the work reaches her desk.
For a strategy, that might mean asking:
What data, inputs, or context should be analyzed?
What specific challenges and opportunities should be outlined?
How does each recommendation connect back to the actual problem?
What are the tradeoffs of each decision? (This is the one most workslop misses—AI gives you clean recommendations with no acknowledgment that choices have costs.)
Define what “good” looks like before assigning the work.
Then use AI to help evaluate it.
The conversation she was dreading gets easier when it stops being about AI and becomes about the work.
It can be: ‘I want you using these tools. But the standard for what leaves your desk is the same as it’s always been—maybe higher, because now you have tools that should make the work better.
And before you send it, ask yourself: would I stake my reputation on every recommendation in this document?
Because whether you mean to or not, you already are.’
—
P.S. I’ve finally convinced myself to take a few days off. My sister and I are heading to Montréal next week, which means the next edition will land May 29th, so I can give myself a chance to catch up on client work and all I’ll miss while I’m away.
I made this poster with ChatGPT’s new Images 2 model, which gives you way more creative control than the previous version. Highly recommend playing with it.
I’ve only been to Montréal once, over eight years ago. If you have favorite restaurants, bars, or spots you think would be my vibe, please hit reply and send them my way. I’d be genuinely grateful for any recommendations.
P.P.S. I went to my UCLA Anderson reunion last weekend. It was good to see everyone.
I was genuinely surprised by how many people across completely different industries pulled me into AI conversations with real curiosity and smart questions.
Something has shifted.
People have gone from “I should probably pay attention to this” to “I need to figure this out NOW”.
Not just more clarity on the widening skill gaps, but real desire and urgency to up their own game.
That felt so exciting.
📈 Work with Me
AI Advising and Consulting: Strategic guidance to build sustainable AI strategies and adapt to shifting audience behavior
AI Training: Practical training that helps teams use AI more strategically, effectively, and safely—so it becomes a trusted thought partner, not just a task assistant
What You Need to Know About AI This Week ⚡
Clickable links appear underlined in emails and in orange in the Substack app.
🧠 ChatGPT’s memory finally has a paper trail.
OpenAI is rolling out memory sources, so ChatGPT can show you which saved memories or past chats it used to personalize its responses.
You can delete or correct the context if something is outdated or no longer relevant.
Paid users may also see files from their library or referenced emails from a connected Gmail account.
The feature is rolling out on web first, with mobile coming soon.
🔒 Important privacy note: if you share a chat with someone, they won’t see your memory sources.
👉 Worth doing right now: Go to Settings > Personalization, scroll down to Memory, click Manage, and delete anything outdated.
The update comes with the launch of GPT-5.5 Instant, the new default model for everyone.
If you’ve been reading me for a while, you’re likely already on the Plus plan ($20/month) and using 5.5 Thinking with extended thinking turned on for everything. So, this launch won’t make a big difference for you.
But for everyone else, it’s a nice upgrade.
🚨 Meta is training AI on its own employees’ work.
The company is tracking U.S. employees’ keystrokes, mouse movements, and screen content on Google, LinkedIn, Wikipedia, Slack, and other work-related tools and sites to train AI agents on real work.
Employees using company laptops can’t opt out, and staff backlash was immediate.
Meta likely won’t be alone. As companies look for better human data to train more capable AI systems, more of them will likely do this too.
📸 ChatGPT’s new image model is surprisingly good at a very practical use case: turning everyday photos into polished professional headshots in one shot.
🎶 New viral AI trend: turning text-thread drama into gospel songs
People are pulling screenshots of their text messages into ChatGPT to turn them into plain text, then feeding the text into Suno as lyrics. The top video has over 20M views.
I don’t care what anybody says, this is art.
🏛️ Google’s Pentagon AI deal blindsided its own employees.
Google reportedly signed a classified deal letting the Pentagon use its AI models for “any lawful governmental purpose”—the same broad language OpenAI and xAI accepted, and Anthropic refused earlier this year.
That caught some employees off guard after senior leaders had repeatedly told them Google wouldn’t cave to the Pentagon’s demands and asked them to trust leadership’s judgment.
And that internal tension has turned into a labor fight.
Google DeepMind workers in London have voted to unionize, hoping to block the lab’s AI models from being used by the U.S. and Israeli militaries.
Their concern is that “any lawful governmental purpose” leaves too much room for uses Google once said it wouldn’t support.
⚖️ Meta’s AI copyright win gets a rematch.
Five major publishers and novelist Scott Turow are suing Meta and Mark Zuckerberg, accusing the company of using millions of copyrighted books and journal articles to train its AI models.
Meta already beat a similar copyright lawsuit from authors last year after a judge said they failed to show its AI would actually hurt the market for books.
It wasn’t enough to argue Meta trained on their work without permission. The authors also needed to prove the AI could replace or compete with the originals in a meaningful way.
This new lawsuit is built around that missing piece: the publishers argue Meta’s AI can imitate author styles, generate copycat books, and summarize copyrighted works in enough detail that some readers may not need to buy the originals.
They also say AI-generated books are already flooding Amazon and competing with human-written books.
That’s the legal question to watch: can publishers prove real market harm this time?
📺 China’s Netflix wants most of its shows made by AI.
iQiyi, the Baidu-backed Chinese streamer, expects AI to create the bulk of its films and shows within five years.
It’s also turning its app and website into more of a social video platform filled mainly with AI-generated content.
That sounds like an AI-slop apocalypse. And maybe it will be.
But the business story is more interesting.
iQiyi is under real pressure: revenue has fallen 13% in Q1, while TikTok-style short-video apps like Douyin keep pulling attention away from traditional streaming.
So the company is trying to solve two problems at once: expensive professional production and the audience shift toward short-form, social video.
That means changing the kind of platform it is: less premium streamer, more AI-powered creator marketplace.
Its new AI production platform can help with scriptwriting, directing, visual design, editing, and more. It also gives creators access to iQiyi’s IP library, talent network, digital assets, and distribution system.
Creators who make AI-generated or shorter serialized shows can get an extra 20% revenue incentive through the end of 2026.
AI can lower the cost of making content. It can help creators make more, faster. It can open the door for weird, niche, experimental formats that would never get traditional funding.
If the result is a flood of mediocre AI content, the platform risks training audiences to value the platform less.
More supply doesn’t fix an attention problem.
It can (and will) make it worse.
🤝 Dario gets into bed with Elon.
Anthropic just formed one of the wildest AI alliances we’ve seen: a deal with SpaceX to rent compute from Elon Musk’s massive AI data center in Memphis.
Compute is the infrastructure behind AI: chips, power, servers, cooling, and the capacity to train, improve, and run these models at scale.
Without enough of it, even a great product can’t keep up with demand, which means tighter usage limits, slower service, and frustrated paying users.
So yes, Anthropic—the values-and-safety AI company—is now relying on Elon’s infrastructure to give Claude’s paying users more capacity.
The same Elon who just months ago was mocking Anthropic, calling it “MisAnthropic,” saying it “hates Western Civilization,” and predicting it could never win.
Now he gets a serious customer for his massive AI buildout.
Dario gets compute.
And Sam gets the message: Dario isn’t too principled for AI’s “whatever it takes to win” era.
It is an official cage match. The gloves are off.
And we’re all MMA style now, motherfuckers.
🏆 An AI actor can’t win an Oscar. Yet.
The Academy’s new rules say acting nominees must be credited and “demonstrably performed by humans with their consent,” while screenplays must be human-authored.
The rules don’t ban AI from films, but they draw a line around awards eligibility: AI can be part of the process, but the acting and writing prizes still need a human behind the work.
In case you missed the last edition, you can find it 👇:
🤓 Claude Just Got Smarter—And Pickier About How You Talk to It
It has been an insane couple of weeks in AI.
That's all for this week. See you in 3 weeks.
Thoughts, feedback and questions are always welcome and much appreciated. Shoot me a note at avi@joinsavvyavi.com.
Stay curious,
Avi
💙💙💙 P.S. A huge thank you to my paid subscribers and those of you who share this newsletter with curious friends and coworkers. It takes me about 20+ hours to research, curate, simplify the complex, and write this newsletter. So, your support means the world to me, as it helps me make this process sustainable (almost 😄).