Microsoft’s AI Sales Hit a Wall

Microsoft has spent billions convincing enterprises that AI agents are the future of work. This week, internal reports suggest customers aren’t buying it as fast as projected. Meanwhile, deepfake attacks have gotten so good that a new startup just raised $28 million to help you verify your coworkers are actually who they say they are.

📰 The Rundown

💼 Microsoft’s AI Sales Hit a Wall of Enterprise Skepticism

➡️ The move: Multiple Microsoft divisions quietly lowered growth targets for AI products after sales teams missed aggressive goals, according to The Information. Less than 20% of one Azure sales unit hit its 50% growth target for Azure AI Foundry, the platform for building AI agents. Microsoft publicly denies lowering “quotas” but acknowledged adjusting “growth targets,” a distinction that did little to calm investors. The stock dropped nearly 3% on the news.

⚡ Why it matters: This is the first crack in the AI hype cycle’s armor. An MIT study found only 5% of AI projects advance beyond pilot stage. The Carlyle Group reportedly scaled back Copilot Studio spending after struggling to get the tool to reliably pull data from other applications. Enterprise customers want AI that works, not AI that mostly works.

🎯 Your takeaway: If your company is piloting AI tools that aren’t delivering clear value, you’re not alone. The lesson here: start with specific problems, not general AI adoption. The companies succeeding with AI aren’t chasing every new feature. They’re finding the one or two use cases that actually work.

🎭 $28M Bet: Your Next Video Call Might Be a Deepfake

➡️ The move: Israeli cybersecurity startup imper.ai launched from stealth with $28 million from Redpoint Ventures and Battery Ventures. Founded by veterans of Israel’s elite Unit 8200, the company detects deepfakes and voice clones in real time across Zoom, Teams, Slack, and other communication platforms. Unlike tools that scan for AI-generated content, imper.ai analyzes device telemetry, network signals, and behavioral patterns that attackers cannot fake.

⚡ Why it matters: The Identity Theft Resource Center reports a 148% surge in impersonation scams over the past year. The FBI recorded $2.8 billion in business email compromise losses last year alone. Deloitte estimates AI-driven impersonation could cost U.S. companies $40 billion annually by 2027. A recent attack using fake credentials at Jaguar Land Rover led to an estimated $1.5 billion in losses.

🎯 Your takeaway: Deepfakes have moved from celebrity face-swaps to corporate fraud. When someone urgent messages you asking for a password reset or wire transfer, pause. Verify through a different channel. The most sophisticated AI attack surface isn’t your network. It’s your trust.

⚖️ Legal AI Hits $8 Billion Valuation as Law Firms Go All In

➡️ The move: Harvey raised $160 million led by Andreessen Horowitz, valuing the legal AI startup at $8 billion, up from $5 billion in June. This follows a $300 million Series E earlier this year. Harvey now counts over 50 of the top 100 law firms as customers, using its large language models for legal research and document drafting.

⚡ Why it matters: Legal services represent one of the clearest AI adoption success stories. Unlike general enterprise tools struggling for adoption, Harvey targets a specific industry with clear, high-value problems: legal research takes hours, documents need precise language, and billable time is expensive. The lesson is focus. While Microsoft battles to prove AI agents work everywhere, Harvey proves they work somewhere.

🎯 Your takeaway: Watch which industries adopt AI fastest. Legal, accounting, and customer service are leading because they have clear, repetitive tasks where AI reduces time without requiring perfection. If your field is lagging, look at what those industries are doing right.

🔧 Tool Spotlight: Descript

Descript turns audio and video editing into text editing. Upload a recording, and Descript generates a transcript. Edit the text, and it edits your media. Delete a sentence, and the audio cuts. Fix a typo in the transcript, and the audio stays clean.

What makes it different: Most transcription tools give you text. Descript gives you control. The “Overdub” feature lets you add words to your recording using AI voice cloning of your own voice. Studio Sound removes background noise and echo. Filler word removal automatically cuts your “ums” and “ahs.”

Best for: Podcasters, video creators, and anyone who repurposes meeting recordings into content. But it’s also powerful for cleaning up client presentations or creating training materials from recorded calls.

Pricing: Free tier offers 1 hour of transcription monthly. Creator plan ($15/month) includes 10 hours. Pro ($30/month) adds 30 hours and premium features like Studio Sound.

👉 Try it: descript.com

✨ Try This Today: The Verification Pause

The next time you receive an urgent request, whether via email, Slack, text, or video call, pause before acting.

The technique: Before responding to any request involving credentials, money, or sensitive access, verify the sender through a different communication channel than the one they used to contact you.

How to do it:
1. Receive urgent request (password reset, wire transfer, access approval)
2. Do not reply in the same channel
3. Call the person directly at a number you already have on file
4. Or message them on a different platform: if they emailed, Slack them. If they Slacked, text them.
5. Confirm the request is real before acting

Why it works: Deepfakes can clone faces and voices, but attackers cannot hijack every communication channel simultaneously. A 30 second verification call defeats even the most sophisticated impersonation.

Real example: An attacker pretending to be your CFO on a video call asks you to approve an urgent wire transfer. You say, “Let me pull up the details. I’ll call you right back.” You call the CFO’s direct line. They have no idea what you’re talking about. You just stopped a fraud attempt.

Time required: 30 seconds to 2 minutes. Potential savings: everything.

✨ The Wire

🔗 Werner Vogels delivered his final AWS re:Invent keynote after 14 years, addressing developer fears about AI taking jobs while announcing Trainium3 chips with 4x performance gains. TechCrunch

🔗 OpenAI’s investor-customer flywheel is tightening, with SoftBank building data centers while using OpenAI’s models, and Thrive Capital forming AI “roll-ups” that embed OpenAI researchers directly with domain experts. Reuters

🔗 The EU opened an antitrust probe into Meta’s WhatsApp AI policies, investigating whether the company unfairly restricts third-party AI providers and advantages its own services. CNBC

Neural Notes — AI that amplifies your value, not replaces it.