This Sunday is about how we work with AI: what still matters when you write to a chatbot, how AI answers are changing Google Search, and why running many coding agents at once can feel more productive than it is. Here are six stories, plus one surprise.
How we checked: we read Forward Future’s daily newsletter of October 30, 2026 (the page could only be partly loaded) and the public news summaries on X from the last 36 hours. The X summaries are written by an AI from many posts; we did not check them against original pages, so facts from them are labelled “according to an X news summary”. We did not test any product.
Key terms:
- Prompt: the instruction or question you write to an AI.
- Prompt engineering: tricks and fixed phrases that people used to get better answers from an AI, such as “Think step by step”.
- AI agent: an AI program that completes a task on its own in several steps, using tools along the way.
- Open source: the program code is public and may be used and changed for free (here under the Apache-2.0 license).
- Code review: a second look at newly written code to find mistakes before it goes live.
- Open weights: the finished model is released for download, so anyone can run it on their own hardware.
1. Prompting without magic words: what really matters now
- What changed: Forward Future asked its own team whether “prompt engineering” still matters. The answer was surprisingly clear: old tricks like “You are an expert in …” or “Think step by step” barely help with today’s models.
- Quotes: founder Matthew Berman calls classic prompt engineering “basically useless”. Editor Brian Knepprath says spending a long time polishing a prompt is now a waste of time.
- What matters instead: co-founder Nick Wentz says you should no longer tell the AI HOW to do something, but WHAT you want, WHY it matters and WHAT a good result looks like. Good AI users explain the problem clearly, give the right background and recognize a good result when they see it.
Original article: Forward Future, Oct 30, 2026
2. Google: AI answers bring more searches, but investors are skeptical
- The numbers: according to Forward Future, quarterly revenue at Alphabet (Google’s parent company) rose 25 percent to 127.6 billion dollars, but stayed below what the stock market expected. Revenue from Google Search grew 12 percent to 63.35 billion dollars.
- What it means: since Google added AI answers to Search, people search more and ask longer, more complex questions. The open question is whether Google can sell ads on such questions as well as it can on classic searches.
- More headlines from the same issue (headlines only, we could not load the details): Chinese AI stocks lost 480 billion dollars in value, and NVIDIA is under pressure from a spy affair and competition investigations.
Original articles: Forward Future, Oct 30, 2026 · Forward Future newsletter archive
3. Coding with agents: real progress or just a rush?
- The warning: according to an X news summary, former Anthropic developer Thariq Shihipar describes a kind of “mania”: people who run more and more AI agents at the same time feel very productive, but often deliver little that is finished. One person reported 10 hours of overnight agent work with only a few usable results. His advice: measure success by finished features and fixed bugs, not by the amount of code.
- The other side: according to a second X news summary, agents now deliver code like a constant stream. At Anthropic, daily code changes are said to have grown from 2,500 to more than 30,000, which is why the company reportedly built its own system to manage code.
- Related: according to an X news summary, Ruby on Rails creator David Heinemeier Hansson (DHH) thinks AI is even better at checking code (code review) than at writing it.
Original articles: X News: Developers Hooked on AI Coding Agents Without Real Progress · X News: AI Agents Reshape Software Engineering into Continuous Code Streams · X News: DHH Challenges Developers on AI Code Review Hype
4. Aura Autopilot: a free AI that plays games by itself and finds bugs
- What’s new: according to an X news summary, Aura (from the company Ramen) has released its “Autopilot” tool as open source on GitHub, under the free Apache-2.0 license.
- What it does: the agent works in the game engines Unity, Unreal Engine and Godot. It reads the game’s code, plays the game in real time and follows simple instructions like “defeat 10 enemies”. It reports bugs with proof, for example an elevator that won’t go down.
- How good it is: according to the summary, it is 43 percent more accurate and almost twice as fast as Claude Code at this task. That mainly helps small teams without their own testers. These numbers come from the vendor and have not been independently checked.
Original article: X News: Aura Open-Sources Autopilot AI for Smarter Game Bug Hunting
5. AI video: Pika H3 Flex and a model agency without real models
- What’s new: according to an X news summary, Pika H3 Flex creates 15-second videos with matching sound in about 7 seconds, in up to 4K quality. It is said to be 1.8 times faster and 80 percent cheaper than the previous version and to top the ranking of open-weights video models. These are vendor claims.
- Handy: you can add up to 12 reference images and get speech, music and sound effects as separate audio tracks.
- The downside: according to the summary, the company Glif has launched the first “AI model agency”, which turns photos into artificial models and uses them to create whole ad videos. Critics see jobs for models, photographers and film crews at risk.
Original article: X News: Pika Labs Releases Pika H3 Flex for Fast AI Video with Audio
6. Gemini 4 Argon: Google’s new model for complex tasks
- What it is: according to an X news summary, Google introduced “Gemini 4 Argon” at the end of October. It breaks big tasks into steps and creates, for example, web apps, 3D files or finished reports.
- New features: prototypes can be built in Google Antigravity, PDFs and Google Docs can be edited right in the chat, and there are connections to tools like Linear (a task tracker for software teams).
- Limits: developers praise its clear explanations, but some notice occasional mistakes with architecture details compared with rival models. This means software architecture, the way a program is structured, not buildings.
Original article: X News: Google’s Gemini 4 Argon Tackles Complex Tasks for Developers
Surprise of the day: a whole computer processor hidden inside an AI model
- According to an X news summary, developer Divyansh Agrawal has built a computing core called “QwenPU” directly into the weights (the learned numbers) of the open model Qwen3.5-4B. Before that, he had already fitted a complete 32-bit processor into it, with layers of the model serving as working memory.
- What it does: the model can run small C and MicroPython programs during its normal computation, which makes it much more accurate at math where it used to make mistakes. A crazy but exciting experiment.
Original article: X News: QwenPU Embeds CPU in AI Model, Revives Accountability Debate
Sources
- Forward Future daily newsletter, Oct 30, 2026
- Forward Future newsletter archive
- X News: Developers Hooked on AI Coding Agents Without Real Progress
- X News: AI Agents Reshape Software Engineering into Continuous Code Streams
- X News: DHH Challenges Developers on AI Code Review Hype
- X News: Aura Open-Sources Autopilot AI for Smarter Game Bug Hunting
- X News: Pika Labs Releases Pika H3 Flex for Fast AI Video with Audio
- X News: Google’s Gemini 4 Argon Tackles Complex Tasks for Developers
- X News: QwenPU Embeds CPU in AI Model, Revives Accountability Debate
