The agent layer arrived this spring. Here's what Central Coast businesses should actually do about it.
For two years we have lived in a world where AI meant chat. You opened a tab, you typed, it answered. Useful, but a tool you had to remember to use. In the past 90 days that quietly changed. Anthropic, OpenAI, Notion, and the open-source community all shipped serious agent products within weeks of each other. The shape of how businesses use AI is shifting from a thing you talk to into a thing that runs in the background and does work.
This is the moment that matters for a Central Coast business. The tools just got mature enough that you can put one agent on email triage, another on proposal drafts, another on your Monday morning briefing — and they will actually do the job rather than need a babysitter. Below is the field map: what shipped, what it actually does, and what to do about it before competitors who are already running these systems eat your lunch.
The four launches that matter
Anthropic shipped Plan Mode and Claude Design
Plan Mode is the feature that turned Claude Code from a fast typist into a colleague who reads the problem before writing. You ask Claude to do something non-trivial, it explores the codebase, identifies what is affected, and proposes the plan before touching anything. For businesses, the equivalent is asking Claude to draft a customer reply and watching it pull the right context from your knowledge base, your CRM, and the customer's history before composing — not just typing words.
Claude Design landed in April. It is an AI tool that takes prompts and produces real prototypes, slides, one-pagers, and visuals — and it learns your brand by reading your codebase and design files first. For owners who do not have a designer on staff, this is a meaningful unlock. Deep-dive: see our breakdown of Plan Mode and Claude Design.
OpenAI shipped AgentKit
AgentKit is OpenAI's answer to the question many small businesses asked first: how do non-engineers build agents? The Agent Builder is a drag-and-drop visual canvas. The Connector Registry is a managed pool of integrations — Dropbox, Google Drive, Sharepoint, Microsoft Teams, and a growing list of third-party MCP servers. ChatKit lets you embed the agent into your own product. ChatKit and Evals are generally available; Agent Builder is in beta. Deep-dive: see our walkthrough of AgentKit.
Notion shipped Custom Agents
If you run your business in Notion — and a surprising number of Central Coast firms do — Custom Agents may be the highest-leverage thing on this list. Launched in February 2026, in two months teams created more than half a million of them. They run on schedules or triggers, take a job description in plain English, and just work. In April they got their own Plan Mode and a code execution environment called Workers for Agents. Costs dropped 35 to 50 percent in May. Deep-dive: see our guide to Notion Custom Agents.
The open-source world shipped Hermes
Hermes Agent from Nous Research crossed 60,000 GitHub stars in two months. It is an open-source agent that lives on your server, plugs into Telegram, Discord, Slack, WhatsApp, Email, and CLI, and — critically — has a built-in learning loop. It builds skills from experience and persists what it learns about your business across sessions. For businesses that cannot send data to a US tech giant, this is the option that exists now rather than next year. Deep-dive: see our look at Hermes Agent.
The pattern across all four
Four products from four very different teams, but the same four design moves appear in all of them:
One — plan before act. Every one of these agents now thinks first. Plan Mode in Claude. Plan Mode in Notion. A planning step is the new baseline because the old habit of asking AI to just do it turned out to cause expensive mistakes.
Two — persistent memory. Hermes does it natively. Custom Agents in Notion remember the context of the workspace. Claude has Skills and Projects. The era of starting from scratch every conversation is over.
Three — no-code building. AgentKit's Agent Builder, Notion's plain-language agent creator, Claude Design's prompt-to-prototype flow. Building an agent in 2026 looks more like writing a job description than writing code.
Four — embed anywhere. ChatKit puts the agent in your product. Hermes connects to whatever messaging platform you live in. Custom Agents trigger off any Notion event. The agent is no longer a destination you visit. It comes to where you already are.
What this means if you run a business in Santa Barbara
The combined effect of these four launches is that the cost-and-complexity bar for adopting an agent dropped about a full level in 90 days. A year ago, putting an AI agent on your email triage required hiring a developer or buying a vertical SaaS for a specific industry. Today, you can build it in Notion in an afternoon, embed it in your existing tools via AgentKit, or self-host it on a $20-a-month VPS with Hermes. The decision is no longer whether — it is which stack.
What we are seeing locally: the businesses that have already adopted an agent for one repetitive task — inbox triage, proposal drafting, Monday-morning reports — are reclaiming six to twelve hours of owner time per week. That compounds. Twelve hours a week back is more than half a workday. Across a quarter, that is the time to land two new clients you would not have had the bandwidth for.
A 90-day plan that will not burn you
If you are sitting at zero today, the path looks like this. Days 1–14: pick one repetitive task — the one your team groans about most. Days 15–30: build the agent for that task in whichever of these four products fits your existing stack (Notion if you live in Notion; AgentKit if you live in a custom app; Claude Skills if you live in Google Workspace; Hermes if you have strict data residency needs). Days 31–60: measure. How many hours did it actually save? Did it make mistakes that needed cleanup? Tune. Days 61–90: pick the second task. Now you have a system, not a stunt.
The trap to avoid is buying a platform before you have shipped a single agent. Pick one task, ship one agent, measure one result. Then expand.
Where this is going next
The next 90 days will bring a deployment surface for ChatGPT agents, a generally-available Notion Workers for Agents, and a fight over which agent gets to be the default in Microsoft, Google, and Apple operating systems. None of that should change your move right now, which is to ship one agent against one specific task. Everything after that is easier once you have done it once.
If you want help picking the first task — the one that actually moves your week — we do a free 30-minute call. No slides, no sales pitch. We come with a list of common quick wins for businesses your size, and we leave you with a plan whether we work together or not.