Top story
Microsoft Merges Its Copilot Apps and Wipes Group Chats, Deep Research, and More
Source TechCrunch
Why it matters for entrepreneurs: Microsoft folded its consumer Copilot app and its Microsoft 365 Copilot app into a single "Microsoft Copilot," and by August 18 it permanently deleted Group Chats (with all shared messages and images), AI-generated podcasts, Copilot Labs experimental features, and Deep Research. If you or a client built a workflow around any of those — shared team chats, research summaries, generated podcast recaps — that workflow is gone, not deprecated-with-notice. Files from the standalone app got migrated to OneDrive automatically, but anything living inside the killed features did not survive the cutover.
This is the practical risk of building on rented AI infrastructure: features you depend on can disappear on a vendor's timeline, not yours. Microsoft's own EVP over Copilot said the app needed to "earn the right to exist," which is candid but doesn't help you if you didn't back up your data before the deadline. Worth an audit this week: which client-facing workflows sit inside a single vendor's UI with no export path, and which ones route through infrastructure you actually control (Make, GHL, your own database)?
Quick hits
Anthropic's CEO Calls the AI Backlash a "Crisis of Trust"
Source TechCrunch
Dario Amodei said negative public sentiment toward AI is now a serious problem for the industry — and a separate TechCrunch piece the same week found 52% of Americans are "more concerned than excited" about AI in daily life, up from 37% in 2021. If your clients or prospects are in that 52%, leading a sales conversation with "trust and control" beats leading with "capability."
OpenAI Expands Its Cyber-Defense Model as AI-Written Attacks Increase
Source TechCrunch
OpenAI expanded Daybreak, its cyber defense service, into two tiers — Blue (incident response, malware analysis, patch validation) and Red (purpose-trained models for security testing) — and shipped GPT-5.6-Cyber, which OpenAI says clears 95% of advanced cyber test requests. The plain-English version: both attackers and defenders now have AI-grade tooling. If you handle client data (CRM records, health intake forms, financial docs), this is a good week to confirm your own patching and monitoring hasn't fallen behind what the offensive side can now automate.
Tool / launch watch
OpenAI cut GPT-5.6 Luna pricing 80%, to $0.20 per million input tokens and $1.20 per million output tokens, funded by a 20% reduction in its own inference costs. GPT-5.6 Terra dropped 20% to $2/$12. For anyone running high-volume agent workflows — lead scoring, call summarization, intake triage — Luna at this price point makes tasks that were previously too expensive to automate worth running through an LLM again. Worth re-pricing any workflow you shelved earlier this year for cost reasons.
Funding / M&A pulse
- Skan AI raised a $63M Series C (Cathay Innovation and Dell Technologies Capital co-led, with Citi Ventures and Bloomberg Beta participating) to build what it calls a "context layer" for enterprise AI agents — software that watches how employees actually do their jobs and feeds that record to agents. Total raised: roughly $120M since 2019. The signal: investors are betting the bottleneck for enterprise AI agents isn't the model, it's knowing what the actual workflow looks like before you automate it — the same diagnosis SyncBroadAI leads with on every client engagement.
Angle for the blog
The Tech Digest is compiled each morning by SyncBroad AI — a plain-English read on AI for service businesses. Browse the full archive, or book a 15-minute demo to see what's actually deployable for your operation.