This week, four stories delivered the same uncomfortable truth: AI isn’t coming for the workforce. It’s already operating inside it.
Uber
Dara Khosrowshahi revealed that 90% of engineers use AI daily. The real signal? The top 30% are shipping the most code diffs — with AI. His 5-year prediction: the ROI of AI agents + GPUs will surpass a single human engineer.
Hire fewer engineers. Deploy more agents.
AT&T
8 billion tokens per day broke their old stack. They rebuilt around a multi-agent system — LLM “super-agents” directing task-specific workers. Result: 90% cost reduction.
A 6-week data workflow now takes 20 minutes.
ServiceNow
Its Autonomous Workforce now resolves 90%+ of internal IT requests. Tickets close 99% faster than human handling.
Union.ai:Raised $38.1M to build the orchestration layer making this scalable.
The engineer isn't being replaced. They're being promoted. The question is whether you'll make that move — or wait until it's made for you.
Executive Brief
• Uber CEO: 90% of engineers use AI daily — Top 30% are power users reshaping the codebase architecture. Dara predicts: in 5 years, ROI of AI agents + GPUs surpasses ROI of human engineers. The era of hiring fewer devs begins. Read More
• AT&T: 8 billion tokens a day forced a complete AI rethink — Built a multi-agent LangChain stack: LLM super-agents + task-specific workers. A 6-week data project now deploys in 20 minutes. 90% cost cut. Read More
• ServiceNow Autonomous Workforce launches — 90% of internal IT requests handled autonomously. Resolution is 99% faster than human agents. Product goes GA second half of 2026. Read More
• Union.ai closes $38.1M Series A — Bellevue, WA-based AI workflow infrastructure startup backed by NEA, Mozilla Ventures. Supports open-source Flyte orchestrator. Scaling AI from experiment to production. Read More
• Google Gemini 3.1 Pro drops — ARC-AGI-2 score of 77.1% — double the reasoning of 3 Pro. Rolling out to Gemini app, NotebookLM, Vertex AI, and Google AI Studio. Read More
• Claude Sonnet 4.6 is now default for all users — Enterprise-grade AI at accessible pricing. 1M token context window. Improved coding, computer-use, and complex data handling. Read More
• Uber engineers built 'Dara AI' chatbot — Teams simulate CEO feedback before high-stakes presentations. The bot preps slides, predicts objections, and refines proposals before Dara ever sees them. Read More
• Klarna: SaaS is dead in the agentic era — CEO Sebastian Siemiatkowski: AI agents will dissolve SaaS switching costs. Klarna cut from 7,000+ to under 3,000 employees. Plans to cut to 2,000 by 2030. Read More
The engineer isn't disappearing. The engineer is becoming the architect. The question is whether you'll build the blueprint — or wait for someone else to hand it to you.
DEMO THEATER:
AT&T's 8B Token Pivot — How Enterprise AI Actually Scales.AT&T didn’t just scale AI. It had to rebuild for it.
At 8 billion tokens a day across 100,000+ employees, you can’t solve the problem with more GPUs. You redesign the architecture.
CDO Andy Markus rebuilt Ask AT&T as a multi-agent LangChain stack — one LLM “super-agent” routing tasks to smaller, specialist worker agents.
The results:
• 6-week data projects → 20 minutes
• 90% cost reduction
• Near-zero latency
Before: One giant model does everything. Expensive. Slow. Fragile.
Now: A routing layer assigns tasks to specialists. Models are modular. Swappable. Future-proof.
For enterprise builders:
• Multi-agent orchestration isn’t optional at scale. It’s survival.
• Interchangeable models = pricing and performance leverage.
• The bottleneck was never the model. It was orchestration.
It’s what happens when AI stops being a pilot
—and becomes infrastructure.
Builder moves that actually matter
• AT&T + LangChain multi-agent stack: If you're building internal AI tools, stop using one general model for everything. Route tasks. Use specialists. Cut costs 90% in the process.
• ServiceNow Autonomous Workforce: The IT desk just automated itself. If your team still handles Level 1 support tickets manually, this is your signal to evaluate AI specialists in that workflow.
• Anthropic acquires Vercept: Claude's computer-use score jumped from 15% to 72.5% on OSWorld. AI agents that operate actual desktop apps — spreadsheets, browsers, forms — are now viable. Start scoping those workflows.
• Perplexity Computer: 19 models, orchestrated. $200/month. It researches, codes, deploys. The fully autonomous digital worker is no longer theoretical.
• Klarna's workforce trajectory: From 7,000+ to under 3,000 employees. Targeting 2,000 by 2030. Not layoffs — attrition. They're not replacing humans. They're just not replacing roles when humans leave.
The winning move isn't the biggest model. It's the right model, in the right workflow, removing the right step.
MONEY PULSE
Union.ai (Bellevue, WA) → $38.1M Series A. NEA + Mozilla Ventures. AI orchestration infrastructure. Powers open-source Flyte. Scaling AI from experiment to production.
Vercept (Seattle, WA) → Acquired by Anthropic. $50M previously raised. Team joins Anthropic to build next-gen computer-use agents. OSWorld benchmark: 15% → 72.5%.
Klarna workforce reduction → From 7,000+ to under 3,000. Saved hundreds of millions. CEO projects 2,000 employees by 2030. AI absorbed the work. Per-head compensation rose 50%.
ServiceNow + Moveworks integration → $1.5B Moveworks acquisition now fully embedded in the Now Platform. Autonomous Workforce product goes GA in Q2 2026.
Perplexity Computer → $200/month for Max subscribers. 19-model orchestration. The agentic computing race has a commercial offering. Priced to replace a junior freelancer.
AI isn't reducing headcount. It's letting headcount reduce naturally — and then doing the work anyway.
Cahn's 2 Cents: The Quiet Restructuring
No headline said "mass layoffs." No government policy triggered it. No backlash trended.
But this week, the numbers told the same story:
• Uber: 90% of engineers already using AI daily. Power users reshaping the codebase.
• AT&T: AI stack now handles work that took humans 6 weeks. In 20 minutes.
• ServiceNow: 90%+ of IT support requests resolved by AI. 99% faster.
• Klarna: From 7,000 employees to a target of 2,000 by 2030 — not through cuts, but through not refilling.
This is the quiet restructuring. No drama. No protests. Just math.
The real question every team needs to ask is not: Will AI take my job? It's: Am I the person who architects the AI system — or the task it replaces?
Five tools collapsing real loops right now:
These aren't shiny. They're effective.
AT&T multi-agent stack (LangChain) → 6-week data project → 20 minutes.
ServiceNow Level 1 AI Specialist → IT helpdesk ticket → resolved autonomously in seconds.
Anthropic + Vercept (Claude computer-use) → Manual browser/app task → AI-operated desktop agent.
Perplexity Computer → Complex multi-step project → 19 specialized AI models handling it end-to-end.
Uber's Dara AI chatbot → Pre-meeting prep → AI simulates CEO feedback before the real meeting.
If a tool doesn't remove a step, it's noise.
CAHN'S POV
AI companies aren't just competing on capability anymore.
They're competing on:
• Workforce integration — who owns the daily workflows inside enterprise
• Agent autonomy — who can run end-to-end without human intervention
• Safety boundaries — who can hold their red lines under government pressure (Anthropic vs Pentagon)
• Infrastructure ownership — who controls the compute, data flows, and orchestration layer
This week felt less like AI innovation — and more like AI consolidation. Uber became the blueprint for the AI-integrated workforce. AT&T showed what scale demands. ServiceNow automated the first job that wasn't supposed to be automated. And Anthropic drew a line — against the most powerful military in the world — that no other AI lab has drawn.
The winners in 2026 won't have the best demos. They'll have the deepest roots — in infrastructure, in enterprise workflows, and in the trust that makes them irreplaceable.
Quick Beats
This week: Map your team's top 3 repetitive workflows. Then ask: which of the tools above could remove one step? (AT&T, ServiceNow, Perplexity Computer, Claude computer-use)
Try Perplexity Computer: Pick one multi-step project you've been putting off. Drop it into Perplexity Computer and let the 19-model stack take a first pass. The output might surprise you. (perplexity.ai/computer)
Watch Anthropic vs Pentagon: This is the most important AI governance story of 2026. Follow it. Whoever blinks first will define how AI safety policies apply to government contracts globally.
The winning move isn't the biggest model. It's the right model, in the right workflow, removing the right step.
Fireside Chat
Uber engineers are using AI to simulate their CEO. AT&T is routing 8 billion tokens a day through specialist agents. ServiceNow automated the IT desk. Anthropic is holding its red lines against the Pentagon. Which of these signals matters most for the future of AI in 2026 — workforce transformation, infrastructure scale, autonomous agents, or AI safety governance?
AI PUN
Why did the AI refuse to do stand-up comedy? It kept getting stuck in an infinite loop trying to find a punchline that didn't have any bugs.
That's All Folks: If this changed how you see the AI workforce story — not as a future threat, but as a present restructuring — forward it to one person still treating this as a tomorrow problem.
—
Aditi & Swati - The humans behind Cahn's AI Canvas
This week, the workforce didn't just change. It restructured quietly, company by company, workflow by workflow.
→ Big. Systematic. And already in motion.
Stay Creative. Stay Updated.
Build. Learn. Montize on AI with us: [email protected], @ai.cahn
Edition #42 covered Feb 21 - Feb 27, 2026. All news verified from mainstream sources with direct article links provided.
Disclaimer: The information presented in this newsletter is curated from public sources on the internet. All content is for informational purposes only.
