Welcome back! The era of talking to AI is ending. One tech giant just spent billions to put an autonomous workforce inside your favorite messaging app. While social media gets automated, the office is changing too. The CEO of the world's biggest chip company predicts you will soon be hiring and managing digital humans as if they were real staff. On the other side of the world, a new model from China just claimed the top spot in coding benchmarks days before a massive IPO. The software is getting smarter and the workforce is getting stranger.
In today’s Generative AI Newsletter:
Meta buys Manus to bring AI agents to WhatsApp.
NVIDIA predicts companies will hire digital employees.
Z.ai beats Western coding models just before its IPO.
AiPPT turns your documents and links into slide decks.
Latest Developments
Meta just acquired Manus, a Singapore-based startup that became a viral sensation for its autonomous agent capabilities. The deal is valued at $2B, matching the valuation Manus sought for its latest funding round. This acquisition marks a pivot from conversational chatbots to action-oriented agents that can independently execute multi-step workflows. By absorbing Manus, Meta gains a high-growth engine that reached $100M in annual recurring revenue just eight months after launch.
The execution layer of AI:
Agent Performance: Manus reported a 57.7% score on the complex tier of the GAIA benchmark, outperforming early results from OpenAI’s Deep Research.
Commercial Traction: The startup grew from zero to over $125M in total revenue run rate by late 2025, supported by millions of paying users.
Platform Integration: Meta will keep Manus running independently while weaving its autonomous tools into Facebook, Instagram, and WhatsApp.
National Security: Meta confirmed Manus will sever all Chinese ownership ties and cease operations in China to satisfy regulatory concerns.
2026 will be the year of the agent. While Meta has spent billions on its Llama models, Manus provides the execution layer that allows AI to perform work rather than just generate text. Mark Zuckerberg is betting that the future of social media includes digital employees capable of managing a user’s professional and personal life. As these agents hit billions of phones, the standalone assistant era ends. The age of the autonomous executor has begun.
NVIDIA CEO Jensen Huang predicts a future where companies will hire, onboard, and manage "digital humans" as formal members of the workforce. During a recent interview with Citadel Securities, Huang estimated the market for agentic AI labor could reach trillions of dollars as specialized agents join fields like nursing, law, and marketing. Rather than viewing AI as a static tool, Huang argues that enterprises will treat these agents like employees that must absorb a company’s unique culture and philosophies to be effective.
The architecture of AI labor:
HR for Agents: Huang predicts that IT departments will evolve into the HR department for digital employees, responsible for the orientation and management of non-human staff.
Agentic Density: NVIDIA already employs more cybersecurity AI agents than human security staff, using them to safeguard proprietary data at a scale humans cannot match.
Mixed Workforces: Future enterprises will likely license expert agents from platforms like OpenAI or Harvey while building homegrown bots for internal tribal knowledge.
Economic Displacement: Anthropic CEO Dario Amodei warned that this shift could eliminate 50% of entry-level white-collar jobs by 2027 as agents outperform humans in routine tasks.
The transition to a hybrid workforce represents a sharp transformation in how business is measured and executed. While Salesforce CEO Marc Benioff noted that current executives are the last to lead all-human teams, the real challenge lies in integration. Performance metrics will likely shift from individual output to the ability of a human to "onboard" and direct a fleet of digital workers. We are moving toward a labor market where your value is defined not by what you can do, but by how many digital humans you can effectively lead.
Chinese AI startup Z.ai has released GLM-4.7, a new coding-focused model that has claimed the top spot on major open-source benchmarks. This release arrived just days before the company’s $6.7B initial public offering on the Hong Kong Stock Exchange. With a 73.8% score on SWE-bench Verified, GLM-4.7 is the first model from a Chinese lab to break the 70% threshold on real-world software engineering tasks. The startup, backed by Alibaba and Tencent, is positioning the model as a direct rival to Western leaders like Claude Sonnet 4.5.
The new coding standard:
Reasoning Architecture: The model features Interleaved Thinking, which allows it to reason through logic before every response or tool call.
Agentic Performance: GLM-4.7 reached an open-source high of 84.9 on LiveCodeBench V6, surpassing the scores of both DeepSeek-V3.2 and Kimi K2.
Visual Vibe Coding: A specialized focus on front-end design allows the model to generate modern UI layouts with 91% accuracy in slide and web formatting.
Market Entry: Z.ai is offering the model through a $3 monthly subscription, significantly undercutting the pricing of high-end Western proprietary models.
The timing of this release serves as a powerful warning to global investors ahead of Z.ai's January 8 listing. By open-sourcing the model weights on Hugging Face, the company is following a common Chinese strategy of using open-access dominance to build a developer ecosystem. We are moving toward a period where the performance gap between Western "closed" models and Chinese "open" models is effectively vanishing in specialized fields like software engineering.
AiPPT is a tool that transforms PDFs, Word documents, or webpage links into customizable slide decks. It's beneficial when you have existing content and require a quick way to organize it with clear structures, titles and layouts without beginning from scratch.
Core functions (and how to use them):
Document Import: Upload a Word file or PDF to convert headings into slide titles and paragraphs into bullets for a first draft.
Convert URL to Slides: Paste a blog post or landing page URL to create a deck outline, then trim it to 8–12 slides.
Apply Templates: Select a template to ensure uniformity in fonts, spacing and layout on all slides after generation.
AI Editing: Choose a cluttered slide and request a shorter version with a clearer title to enhance readability at a glance.
Export PPT: Export as PPT, open it in PowerPoint or Google Slides, and do final tweaks like charts, icons, and brand fonts.
Try this yourself:
Pick one real content piece today (a 1–2 page doc, a short PDF or one URL) and generate a 10-slide deck. Before exporting, fix three slides: Make slide 1 a plain-sentence title, cut slide 2 to three verb-led bullets, and add one hard detail on slide 3 (number, price, date or quote). Export as PPT, open it, and replace one slide image with a screenshot or chart from the source.
