- Everyday AI
- Posts
- Ep 712: AI Agent Crash, Software Collapses and Non-Human Economies
Ep 712: AI Agent Crash, Software Collapses and Non-Human Economies
Anthropic pledges to pay grid upgrade costs, OpenAI launches a product built by Codex agents, Anthropic AI safety head steps down warning āthe world is in perilā and more
Everyday AI
February 12, 2026
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
Sup yāall! š
Thousands of hours.
Thatās the amount of time we spend EACH year talking about, researching on, and building with AI.
One of our most popular shows over the years? Our AI prediction and roadmap series. We literally condense thousands of hours of experience into a few shows.
And, TBH, we had WAY more than 26 AI Predictions, so make sure to go repost todayās LinkedIn show if you want access to this insanely valuable bonus guide.
See ya tomorrow for Vol 2.
āļø
Jordan
Outsmart The Future
Today in Everyday AI
8 minute read
š Daily Podcast Episode: AI agents are about to break how software and work actually function as day 1 of our 2026 AI Predictions and Roadmap series kicks off. Give todayās show a watch/read/listen.
šµļøāāļø Fresh Finds: T-Mobile Tests Live AI Call Translation, NotebookLM testing 9 New Infographic Styles, Threads Launches āDear Algoā AI Feature and more. Read on for Fresh Finds.
š Byte Sized Daily AI News: Report: OpenAI has a model to catch leakers, U.S. Gov doubling down on AI, AI Safety Chief says, āWorld is in perilā and more. Read on for Byte Sized News.
šŖ Leverage AI: We spend thousands of hours learning and building around AI each year. Today, we share part 1 of our 2026 AI predictions and roadmap. Keep reading for that!
ā©ļø Donāt miss out: Miss our last newsletter? We covered: ChatGPT Deep Research upgraded to GPT-5.2, Zhipu releases GLM-5, Claude Cowork hits Windows and more Check it here!
Ep 712: AI Agent Crash, Software Collapses and Non-Human Economies
AI developments legit change hourly.
That means what business leaders can do changes daily. And drastically.
As soon as you FINALLY learn a new AI technique, itās often already outdated, and approvals to use it at work can take forevvvvver.
Solution: know whatās coming.
Thatās why you have me.
Iām not a mind reader, but I spend almost all day every day talking with the people building AI, using it myself, and teaching others.
Itās literally my job.
Each year I publish predictions and a roadmap because you donāt have 10 hours a day to keep up.
(I do.)
So, today is part 1 of one of our most important shows of the year: Our AI Predictions and Roadmap series.
Also on the pod today:
ā¢ AI labs buying student data š«
ā¢ Agents hiring agentsānon-human economies š¤
ā¢ Messaging as universal agent control š±
Itāll be worth your 59 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Gro Automates outreach with an AI Sales Agent, Ava makes editing real video as simple as saying what you need, Lindy saves you two hours a day by proactively managing your inbox, meetings, and calendar
Google Gemini Deep Think ā Gemini Deep Think is solving real research problems in math and science, not just student contests. Click here to Learn More
T-Mobile Live Translation ā T-Mobile is testing live AI call translation in 50+ languages. Beta sign-up is open now.
AI and Politics ā Anthropic is throwing $20 million into the political ring to back tougher AI safety rules. The battle over how to regulate artificial intelligence in Congress just got a lot more interesting.
NotebookLM Infographic Customization ā NotebookLM is testing 9 new infographic stylesāincluding anime and 3D clay.
Fitbit AI Health Coach ā Fitbitās health coach is now available in more countries. Check if you can try it out.
AI and Power ā Anthropic says itāll cover the extra power costs from its AI, so your electric bill wonāt spike as data centers grow. Hereās how.
AI Video ā Runwayās $315M raise fuels rapid AI video gains, challenges Google and OpenAI.
Threads āDear Algoā Released ā Threadsā new "Dear Algo" feature lets you instantly customize your feed by posting what you want to see.
Coinbase AI Crypto Wallets ā AI agents can now trade and manage crypto wallets on their own. Curious how bots are running the show?
MemOS OpenClaw Plugin ā MemOSās plugin lets OpenClaw agents share memory, cutting token costs by up to 70 percent.
1. Report: OpenAI Deploys AI Sleuth to Trace Internal Leaksš°
According to The Information, OpenAI has rolled out a custom version of ChatGPT that hunts for internal leaks by analyzing internal documents, Slack, and email messages.
When sensitive internal news goes public, this AI-powered detective scans company data to pinpoint potential sources and suggest who had access. The toolās architecture reportedly allows complex data searches using plain English, plugging directly into company knowledge bases.
2. Engineers Stop Coding: OpenAI's Codex Agents Build Million-Line App in Months š¤
In a striking shift for software development, OpenAIās team revealed they built and deployed a fully functional internal productāone million lines of codeāentirely generated by Codex agents, with humans guiding instead of coding.
Over just five months, engineers moved from writing software themselves to designing environments and feedback loops, letting Codex handle everything from logic to documentation. The experiment signals a dramatic redefinition of what it means to "engineer" software, as agents now manage the bulk of development while humans focus on high-level goals and system design.
3. AI Safety Chief Quits Anthropic, Warns āWorld is in Perilā š¤Æ
In a dramatic move, Anthropicās AI safety lead Mrinank Sharma announced his resignation with a stark warning about global risks, citing crises from AI to bioweapons.
Sharma said heās leaving the high-profile AI startup to study poetry and ābecome invisibleā in the UK, highlighting growing tensions inside leading AI firms as commercial priorities challenge safety principles. The shakeup comes as Anthropic and rival OpenAI spar over ethics and advertising, with critics suggesting both are drifting from their original missions.
4. Google Ads Gets Creative Boost with Veo 3 Integration š¹
Google has rolled out Veo 3 support in its Ads Asset Studio, giving advertisers access to studio-quality creative tools powered by AI. This move is set to speed up ad production and make high-end visuals accessible in just minutes, directly within Google Ads.
The update reflects a surge in generative media use, as marketers embrace AI to scale their campaigns quickly. Google's latest upgrade positions it as a leader in automated, creative advertising solutions at a time when brands demand more agility and impact.
5. Pentagon Pushes for AI on Classified Networks šŖ
According to reports, The Pentagon is fast-tracking negotiations with top AI companies like OpenAI and Anthropic to bring powerful AI tools onto classified military networks, loosening many user restrictions in the process. Officials say this move could reshape military decision-making, but sparks concern over the risks of mistakes in sensitive settings.
OpenAI has already agreed to let the Defense Department use its tools, including ChatGPT, on an unclassified network for millions of employees, with fewer restrictions than usual.
Everything changed in December and January. Not slowly. The entire AI landscape shifted while most companies were still begging for budget approval on last year's tools.
We could've dropped our 2026 predictions in January like everyone else. But every single note from December? Cooked. Completely useless.
(And sorryā¦. but if your company's AI strategy was written before February, it prolly needs surgery too.)
We laid out 13 predictions on today's Everyday AI that map exactly where agents, enterprise, and your career are headed this year. Volume one of two.
Time to capitalize shorties.
1. AI Labs Come for University Data š„
At least one frontier AI lab announces a structured partnership with a university centered on student-generated reasoning data.
The internet is polluted with AI slop. Labs are starving for real human thinking. Classrooms have it.
Universities are going broke. AI labs are sitting on mountains of cash. Perfect storm.
Students will prolly get free tuition out of this. The "research collaboration" framing is just gift wrapping.
2. VC Firms Morph Into Venture Studios ā”
Venture capital firms begin building in-house engineering and taking 30%+ equity instead of writing checks.
Engineering costs just went to basically zero. OpenAI's own engineers admitted new features are 100% model-written.
VCs already have distribution. Now they don't need your engineers either. Why write a check anymore?
Might not fully hit until 2027. But the math is already screaming at anyone paying attention.
3. Grounded Only Mode Goes Mainstream š
A major AI lab ships a dedicated mode that restricts outputs exclusively to user-uploaded data. No training data. No internet.
NotebookLM is growing faster than every other lab's product month over month. The reason? Grounded data only.
Enterprises want a button that says "use my stuff, nothing else." That button doesn't really exist yet.
Either OpenAI or Anthropic ships this with real marketing behind it. It becomes a defacto enterprise standard.
4. Google Wins Short and Long š„
Google dominates benchmarks at will while quietly building the most durable long-term advantage in AI.
When Google drops a model, it wins every benchmark. Every time. Other labs win categories. Google wins all of them.
The long game is YouTube. An ungodly amount of human video that no other lab can touch.
Their Genie 3 world model and robotics came out of nowhere. Google's been playing chess while everyone else plays checkers.
5. Agents Start Hiring Other Agents ā”
Enterprises deploy internal marketplaces where AI agents dynamically delegate tasks to and hire other specialized agents.
Sounds like 2050. It's technically already happening. Agents now have commerce protocols and shared languages.
Your company's agents will go hire another company's agents autonomously. Humans won't always know how the work got done.
Finance departments gonna require cost attribution per agent run. Wild.
6. Software Stocks Take a Beating š„
Traditional software ETFs suffer a sustained drawdown as AI-native startups eat legacy companies alive.
Some software ETFs are already down 24% year to date. It's gonna get worse. Way worse.
Legacy software is a 150-year-old house. Good luck with those pipes. You'd always rather have the new build.
The losers are selling expensive UIs that agents can bypass entirely. Before Deborah even finishes her vendor review.
7. Messaging Becomes the Agent Remote Control š
Text, Slack, and phone calls become the primary interface for controlling AI agents.
General agents flopped in 2025 because the interface was wrong. Nobody wants another dashboard.
People are already texting their OpenClaw agent via iMessage. Their agent calls them when it needs something. Done.
The future ain't a new UI. It's the same UI we've always had. But the other end is an agent.
8. Agent Crash Makes National Headlines ā”
A high-profile autonomous agent misaction triggers national regulatory scrutiny.
Not hacking. Not a bad prompt. An authorized agent deployed intentionally by a real company goes off the rails.
Remember the lawyers copy-pasting hallucinated sources? That ran for years. This is gonna be way bigger.
Companies are racing to go agentic because competitors will 5x output. That rush leads straight to a crash.
9. Shadow AI Triggers a Fortune 500 Breach š„
Unauthorized employee AI usage leads to a public data leak at a Fortune 500 company.
Companies won't train people on AI AND they lock down access. So employees paste secrets into random AI wrappers.
Enterprise leaders are out here asking about sketchy one-person software tools. Nah. Don't do that.
This might get bad enough that some companies temporarily lock down copy-and-paste. (Sorryā¦. but it's true.)
10. Agent Audit Logs Get Mandated ā”
Regulated sectors begin requiring audited logs of every single action an autonomous agent takes.
If an agent denies a loan, you log exactly why. No more black boxes. No more opaqueness.
Governance, guardrails, responsible AI. Those buzzwords stop being buzzwords and start being lifelines for your company.
Do your agents have IDs right now? Who's tracing decisions? Start building this before regulators make you.
11. Skill Marketplaces Become the Biggest Risk š„
A major agent ecosystem gets breached via malicious skill distribution.
The most downloaded skill on OpenClaw's ClawHub? Straight up malware. Over 230 malicious skills uploaded in one week.
It's absurdly easy to weaponize. Make 10,000 industry-specific skills with injected malware. People download them blindly.
This ain't a laptop virus. It's malware powering an autonomous agent making decisions for you while you sleep.
12. Twelve-Hour Autonomous Task Horizon ā”
METR benchmarks show AI completing 12-plus hour human tasks at 50% success rate by year's end.
A year and a half ago the ceiling was 45 minutes. Now it's seven to nine hours. Twelve is coming.
One agent. No supervision. Research competitors, build a scraper, set up a database, debug code, write the memo.
AI moves from tool to functional employee in 2026. Prolly a junior one. But one that never calls in sick.
13. Agent Ops Becomes a Real Job Title š
Large enterprises establish dedicated agent ops teams to monitor, maintain, and optimize thousands of running agents.
Greg Brockman, President of OpenAI, basically said he has FOMO if his agents aren't running 24/7.
Agent ops becomes as normal as DevOps. Large teams just watching thousands of agents. That's the whole gig.
By 2030, we're all the buns of the agent sandwich. Directions on the front end. Feedback on the back end.
Reply