Cheaper fast models, agents that run on a schedule and new rules on labelling AI. A short tour of what changed and what it means for a small team.

It is hard to keep up with AI news when you also have a business to run. Here are the changes from the last few months that we think matter for a small team, with the sources if you want to read further.

1. Capable AI got much cheaper

Anthropic released three new Claude models in three weeks: Opus 5.5 on September 22, Sonnet 5.5 on September 28 and Haiku 5.5 on October 7. The pattern across all three is the same. Each does more for less. Anthropic says Opus 5.5 costs about 40% less to run than Opus 5, and that Haiku 5.5 costs about 75% less than the small model before it.

Why it matters: the everyday jobs that suit AI, like sorting enquiries, summarising a long email thread or pulling details out of a form, are now cheap enough to run on every message rather than a sample.

2. AI that works while you’re away

The big platforms now offer agents that run on a schedule or in the background. On Anthropic’s developer platform, managed agents gained scheduled runs in June and, in October, a beta feature that lets an agent plan and run multi-step workflows. Coding tools moved the same way, with sessions that keep working in the cloud.

Why it matters: “check the inbox every morning and draft replies to the routine questions” is now a realistic small project, not a research exercise. It still needs a person to review the drafts.

3. Agents can use a computer

Anthropic’s computer use tool, which lets Claude look at a screen and click, type and scroll, came out of beta in August, alongside a new browser use tool. This matters for businesses stuck with an old system that has no way to connect to anything else. An agent can, carefully, work the screens a person would.

4. A common plug for AI tools

The Model Context Protocol, the open standard for connecting AI to apps and data, released its largest update since launch on July 28. It is now simpler to host and easier to secure. The protocol is stewarded under the Linux Foundation, with support from most of the large AI companies, so a connection built once can work with many AI tools.

5. New rules about telling people

In Europe, the AI Act’s general application date arrived on August 2, 2026, including transparency rules: people should know when they are talking to an AI, and AI-generated media should be marked. Canada has no equivalent federal AI law in force, but privacy law still applies to anything an AI tool touches.

6. Developers are using it, and checking it

Stack Overflow’s 2026 developer survey found that most developers who use coding assistants use them daily, and that trust is conditional: the most common answer was that people trust AI output when they can easily verify it. That is a healthy rule for any business too.

What we’d do with this

Pick one task that eats an hour a day, is mostly reading or typing, and is easy to check. Try AI there first, with a person approving the result. If it saves real time after a month, expand. If it doesn’t, you’ve lost very little.

Sources

From the DYOL blog. Photo by Milad Fakurian on Unsplash.

A free 20-minute conversation, then an honest assessment. You’ll know what it costs and how long it takes, in writing, at a fixed price.

Or email lloyd.delacruz@outlook.com. We reply within 24 hours.