The Edge.
BY NICHOLAS BORELLIA clearer view of what’s next in AI.
Issue 009

02 / AI News

AI News

Google wants Gemini to know how your business works.

A blue hub connects a document, an envelope, and a calendar.
AI-generated editorial illustration for The Edge.

Google introduced the Gemini agent as a single assistant for work across applications. Its announcement describes an agent that can plan tasks, use connected tools, retain context, and continue working in the cloud. Google also describes shared coworker agents with their own identities and access to the information teams provide.

The detail I’m watching is continuity. Google says the agent carries memory and skills across its working environments and can route tasks to Gemini or Claude models. That moves the conversation beyond which model you pick to how well the whole system understands the work.

It fits this issue’s theory. A connection to your files gives an agent more information. Your business still needs to establish which information is current, what the rules mean, and who can resolve an exception. These are Google’s product claims; the announcement alone doesn’t establish how reliably they hold up in your workflow.

Read Google’s announcement →

Claude Haiku 5.5 makes routine AI work cheaper.

A compact blue cube sorts cream and coral cards into neat stacks.
AI-generated editorial illustration for The Edge.

Anthropic launched Haiku 5.5 for fast, high-volume work such as summarizing, classifying information, and answering focused questions. The company says it is its most capable small model so far.

For API requests with prompts up to 100,000 tokens, the listed prices are $0.10 per million input tokens and $0.50 per million output tokens. Those rates are 90% below Haiku 4.5’s corresponding prices. Longer prompts have different rates, and the cost of a finished task also depends on how many tokens it uses. This is API pricing, not a claim that a Claude subscription is 90% cheaper.

My takeaway: give the small model a well-defined job and test it. Extracting a date or sorting requests may not need the same model as resolving a complicated customer situation. Count the corrections and review time alongside the usage bill.

Read the launch and pricing →

Model watch: a focused update, not a new ranking. Use the same task and source material when comparing models.

Small businesses are building AI. The support is catching up.

Three blue figures support an arch and place its coral keystone.
AI-generated editorial illustration for The Edge.

Pax8’s latest SMB AI Pulse report says 67% of AI-engaged businesses surveyed have built an AI solution that is in regular use. It also reports that 35% discovered unapproved AI tools, workflows, or agents in their operations during the previous six months.

Those findings come from a survey of 1,150 business leaders across the US, UK, Germany, and Sweden. The companies had 5–499 employees; the figures above concern AI-engaged respondents, not all small businesses. Pax8 sells through a cloud and AI marketplace, so this is vendor-sponsored research.

What interests me is the work around the technology. Someone has to maintain what gets built, decide what it may do, and help people use it well. Buying access is one decision. Supporting the people and processes around it is ongoing work.

Read the findings and methodology →

Primary sources checked October 16, 2026. Product performance statements are attributed to their developers. Commentary reflects my interpretation.