· 5 min read · Muhammad Shah
Tech trends for late 2026: agents that work together, cheaper models and an agent-ready web
Every month brings a flood of AI announcements, and most of it is noise if you're building a real product. These are the shifts from autumn 2026 I think are worth acting on, with what I'd do about each one.
1. Agents are starting to work in teams
O'Reilly's October Radar notes that agents are starting to delegate to, and coordinate with, other agents instead of working alone. Coding assistants are the early example, splitting a large change into pieces that run in parallel.
What to do: design AI features as small, well-defined jobs with clear inputs and outputs. Agents, like people, coordinate far better around "draft a reply to this ticket" than around "handle customer support".
2. Agents need their own identity
If an agent acts inside your systems, it shouldn't borrow a person's login or a permanent API key. The direction the industry is moving in is a short-lived identity tied to a revocable certificate, scoped to the job at hand.
What to do: give every integration its own scoped token, log what it does, and make access easy to revoke. That's good security practice today, and it's what agent platforms will expect tomorrow.
3. Websites are becoming agent-ready
WebMCP is a proposed standard that gives websites a small API to register tools AI agents can discover and call. Instead of an agent clicking through your checkout like a confused visitor, your site could offer "search products" or "book a slot" as clear actions.
What to do: you don't need to implement a draft standard yet, but the groundwork is the same as good technical SEO: clean URLs, structured data, fast server-rendered pages and well-defined actions. This site, for example, publishes an llms.txt summary for AI assistants.
4. Model prices keep falling
Anthropic cut prices by 20% for input and output tokens and by 60% for cached reads, and OpenAI released GPT-6 Sol and Luna with 50% lower prices. For major corporate customers, providers also appear to be moving toward outcome-based pricing: paying per completed task rather than per token.
What to do: revisit AI features you shelved because they were too expensive. Cache repeated context such as system prompts and documents, and track cost per useful outcome, not just cost per request.
5. Quality beats volume in AI-written code
Some open source projects are shutting down external pull requests because most of them are now AI-generated. It's a warning for every team: AI makes code cheap to produce and expensive to review.
What to do: keep people accountable for what ships. Require tests with every change, keep pull requests small, and review AI-written code as carefully as anyone else's.
6. Big tech is reshaping teams around AI spending
All of this has to be paid for. In October, Microsoft offered voluntary buyouts to about 60,000 employees as it invests more in AI. I've written separately about what that means for developers and the teams hiring them.
The short version
- Build AI features as small, measurable jobs.
- Give every agent and integration its own scoped, revocable access.
- Make your site easy for machines to understand, not only people.
- Re-run the numbers on AI ideas that were too expensive last year.
- Review AI-generated code like any other code.
Sources
Keep reading
5 min read
AI-assisted grading: instant, personal feedback without sidelining the teacher
How to design LLM grading that students trust: rubric-first prompts, structured scores, calibration with real answers, and a teacher who stays in charge.
5 min read
Google, Microsoft, OpenAI and Anthropic agree on one robots.txt opt-out for AI training
The four biggest AI companies will honour a shared robots.txt signal for AI training. What it is, when each company starts, how it differs from search and AI answers, and what your website should do.