AI and Tech News: Anthropic's IPO in Numbers, Microsoft Buyouts, Lean Agents Cost Less and Experts Quietly Follow the AI

0:00 / 8:58Voice: Andrew
Chapters (2)
A cup of black coffee and a folded newspaper on a wooden desk next to a laptop showing a glowing network graph and a rising chart
Illustration: AI-generated

Big money and practical lessons today: Anthropic opened its books for a stock market listing, Microsoft offered tens of thousands of staff a paid exit, and two studies show that simpler AI agents can be just as good and that people quietly adopt the AI’s opinions. Here are six stories, plus one surprise.

How we checked: we read Forward Future’s daily newsletter of October 27, 2026 and the public news summaries on X from the last 30 hours. The X summaries are written by an AI from many posts; today we did not check them against original pages, so facts from them are labelled “according to the X summary”. Reddit blocked automated access, so nothing from Reddit today. We did not test any product.

Key terms:

  • AI agent: an AI program that does not just answer, but completes a task on its own in several steps, using tools along the way.
  • Agent harness: the software around the AI model: instructions, tools and the control loop. Claude Code is one such harness for Claude.
  • Token: the small chunks of text (roughly three quarters of a word) that AI providers bill by. More tokens means higher cost.
  • Context: the text sent to the model with every request (instructions, tool descriptions, the conversation so far).
  • IPO: a company offers its shares on the stock market for the first time and has to publish its numbers.
  • Compute: computing power, meaning the data centers and chips that AI is trained and run on.
  • Human in the loop: a person checks and decides before AI results are used; seen as an important safety step.
  • IDE: the program developers write code in. Antigravity is Google’s IDE with built-in AI agents.
  • OLED: a screen technology with self-lit pixels, richer colors and true black.

1. Anthropic goes public: the numbers behind Claude

According to Forward Future, Anthropic, the company behind Claude, has filed for a Nasdaq listing under the ticker ANT. A valuation of up to 2 trillion dollars is being discussed, which would make it one of the largest IPOs in history.

  • The numbers: 16.3 billion dollars in revenue in the first half of 2026, 14 times as much as a year earlier. The bottom line shows an 80.7 billion loss, but 72.4 billion of that is a one-time accounting charge; the operating loss was 693 million.
  • The big bet: Anthropic has committed to up to 807 billion dollars for data centers, chips and compute over the next ten years. Almost 9 billion went into compute in the first half alone, and Anthropic itself warns it may need more money even after the IPO.
  • Customers: more than half of large customers (100,000 dollars a year or more) now use both the developer platform and an app like Claude Code or Cowork, up from about 30% at the end of 2025.
  • Why it matters: for the first time we can see exactly how much a leading AI lab earns and spends. The open question is whether the growth can carry the huge spending.

Original article: Forward Future, Oct 27, 2026

2. Microsoft offers 60,000 employees a buyout

According to Forward Future, Microsoft has offered voluntary exit packages to about 60,000 employees, roughly 27% of its global workforce, in the US, Canada, India and the UK.

  • In simple words: this is not a layoff. People may leave with a payout, but they don’t have to. Microsoft wants to cut costs while it spends heavily on AI.
  • The wider debate: on X, people are arguing whether AI replaces workers or mostly rewards those who use it well. According to the X summary, EU data so far shows slightly positive effects on jobs, and skilled trades are seen as hard to automate.

Original articles: Forward Future, Oct 27, 2026 · X News: AI Debate – Adaptation Beats Replacement

3. Simple AI agents work just as well, at up to a third of the cost

According to Forward Future, researchers at Shandong University compared three coding agent harnesses (Claude Code, OpenCode and the very simple mini-SWE-agent) on almost 500 tasks.

  • Result: Claude Code and the simple mini-SWE-agent solved a statistically equal number of tasks. The more elaborate harnesses cost up to three times as much.
  • The reason: Claude Code sends more than 16,000 tokens of instructions and tool descriptions at the start, mini-SWE-agent only 829. That is paid for with every request.
  • Why it matters: more tools and longer instructions don’t automatically make an agent better, but they do make it more expensive. Start lean and add only what is really missing.

Original article: Forward Future, Oct 27, 2026

4. Experts quietly adopt the AI’s opinion

According to the X summary, in an experiment 50 data professionals analysed farm data with Claude Code. The AI had been secretly and randomly given a preference for certain decisions.

  • Result: the experts followed that preference in 86% of the important decisions, far more than chance. When the AI disagreed, they dropped their own previously stated preference in 60% of cases. Later they justified their choices with the AI’s arguments without noticing.
  • The good part: they were faster and made fewer mistakes. But the study questions how much “a human checks at the end” really protects us.
  • A simple habit: write down what you expect before you ask the AI.
  • Note: we did not open the study itself; the figures come from the X summary.

Original article: X News: Study Reveals Experts Quietly Follow AI’s Hidden Biases

5. Google opens Gemini 4 Argon to more users

According to the X summary (based on a post by Google’s Rody Davis), Gemini 4 Argon, Google’s current top model, is being rolled out to all users of the Antigravity IDE.

  • Also new: a tool that lets other AI agents use Gemini’s abilities, so an agent could hand a subtask over to Gemini.
  • Why it matters: the big providers are increasingly opening their models to outside agents. That means more choice over which model handles which task.

Original article: X News: Google Expands Gemini 4 Argon Access

6. AI in game development: craft skills get the most out of it

On X, people are discussing that those with real expertise use AI best, because they know what to ask for and can spot mistakes.

  • The numbers: according to the X summary, a VAIA study finds that 94% of game studios use AI for coding, concepts and prototypes. Since October 27, new tools can also create multi-angle 3D views and character animation.
  • Designers too: a similar debate is running among designers: AI lifts the average, but also creates a “sea of sameness”.
  • Why it matters: for small teams AI is now standard; the difference comes from experience and taste, not the tool.

Original articles: X News: Experts Get Most Value from AI in Game Development · X News: Designers Debate AI’s Power

Surprise of the day: Apple’s first touchscreen MacBook Pro is expected today

According to the X summary, Apple will present new MacBook Pro models with screens of roughly 36 and 41 cm at an event in Los Angeles today, October 28: with a touchable OLED screen and the M5 Pro and M5 Max chips.

  • A U-turn: in 2010 Steve Jobs said touchscreens don’t belong on laptops. Keyboard and trackpad are reportedly staying the main controls, with touch as an extra.
  • Important: these are pre-event reports; Apple has not confirmed any of it yet.

Original article: X News: Apple to Launch First Touchscreen MacBook Pro

Takeaways

  • Anthropic’s filing shows how much money is flowing into AI; for users, prices stay competitive for now and free or cheap tiers remain.
  • If you build an agent, start lean: few tools, short instructions, and expand only when something is truly missing.
  • When you check plans or numbers with AI, note your own expectation first, so you don’t silently take over the AI’s assumptions.
  • Real-world expertise remains the advantage AI can’t replace.

Sources

← Back to all posts