At Google Cloud Next on 22 April 2026, Google announced its official Agent Skills repository — a standardised way for AI agents to load targeted expertise on demand, rather than trying to know everything all at once. It is a real shift in how the industry is thinking about AI agents. It also happens to be exactly what dareena.ai already built.
What Google announced
The announcement centres on a simple problem: AI agents that receive everything they might need on every single interaction get overwhelmed. Context fills up, costs rise, and the agent starts making inconsistent decisions. The technical term is "context bloat." The plain-English translation is that the agent acts like it was handed an entire filing cabinet when it only needed one folder.
Google's answer — and the open standard they are building around — is Agent Skills. As their post puts it:
"Agents load in skill information only as-needed, reducing the risk of context bloat."
— Google Cloud Blog, 22 April 2026
Each skill is a compact, focused file that packages a specific capability. An agent starts a conversation knowing very little. When the caller's intent matches a skill, the relevant instructions load in. When the skill is done, the agent moves on. The open format for this lives at agentskills.io, and it is now backed by 40+ platforms including Gemini, GitHub Copilot, and Cursor.
What this means in plain English
An AI agent doesn't arrive knowing your business. It knows how to hold a conversation; it doesn't know that your warranty on certain appliance models runs 18 months, or that you need a serial number before you can book anything in.
One option is to train the agent with everything up front and keep it all in memory on every call. That gets expensive quickly, and when the context runs out, the agent starts filling gaps with guesses.
The better option is a library of focused skills the agent reaches for only when the situation calls for it. A sparkie doesn't carry every tool in their pocket. They grab the right one when the job requires it. That is the architecture Google is now standardising across its entire ecosystem — and it is the architecture at the heart of dareena.ai's skills system.
How dareena.ai does it
The dareena.ai skills system has four components that work together.
Agent Skills
Focused behaviours that are always-on, or that trigger when a caller's intent matches. An appliance repair business, for example, might build a Warranty Check skill: the moment a caller mentions warranty, the agent shifts into a specific mode and asks for the model number, purchase date, and serial number before anything else. The caller gets handled correctly every time. The business gets clean, consistent data.
Skill Acquisition Chatbot
The way that skill gets built is not a config file. Not YAML. Not Markdown. The business owner talks to our AI in plain English — "when someone calls about a blocked drain, I need the address, whether it's inside or outside, and how urgent they think it is" — and the system builds the skill from the conversation. The plumber never sees the underlying file. They walk away with a working skill.
Capture Schemas
Structured field definitions woven into each skill. When the Warranty Check runs, the data comes back clean: model number in one field, purchase date in another, serial number in a third — ready to drop into a spreadsheet, log to a CRM, or fire as a webhook. Not a block of text you have to parse by hand later.
Completion Actions
What happens when a skill finishes. SMS the on-call technician. Add a row to Google Sheets. Create a Trello card. Book the appointment in Google Calendar. Send an email. Fire a webhook. These can be chained. You configure them in the portal, not in code.
Why this matters
When Google, Anthropic (via agentskills.io), and dareena.ai all independently arrive at the same architecture, the pattern is probably right. This is not a case of following the leader — the dareena.ai skills system was designed for the specific constraints of NZ small business calls, not for cloud infrastructure management. The convergence is the point.
The difference is that dareena.ai's version is built and ready — not a keynote demo or a repository of cloud infrastructure examples. The Warranty Check example is not a hypothetical. It is a pattern the platform is designed to run from day one, and it is exactly the kind of thing the first customers on the free trial can build today.
And because the Skill Acquisition Chatbot handles the building, none of this requires a developer. You describe what you want in plain English. The system does the rest. The appliance repair business owner is not writing skill files — they are just telling the AI what they need, and getting on with their day.
The bottom line
If you have been weighing up whether AI agents with targeted skills are the right architecture for your business, the industry has now made its call. Google made theirs on 22 April. dareena.ai made ours a bit earlier.
Sometimes the right architecture is obvious in hindsight. We just happened to get there first.