The next phase of AI in B2B

AI Sparkles, Embedded Assistants, and AI Employees

AI is increasingly part of the B2B products I use. I’ll first describe two AI modalities that are now common — what I call “AI Sparkles” and Embedded Assistants. Then I’ll flush out what might be next: AI Employees.¹

AI Sparkles

AI Sparkles are tiny bits of AI-driven polish — small enhancements that don’t change the product’s core but make some part of it easier, faster, or more delightful. They’re subtle — maybe a smart autocomplete here, a better suggestion there. No single AI Sparkle will blow you away, but they add up, and in aggregate they can make a real difference to the user experience.

Three examples:

1. In Linear, if I have Slack connected, I can highlight a message and pick “New issue from Linear.” The issue title and body get filled out automatically.

2. In Arc, tab names and downloads are automatically renamed and tidied up.

   

3. In Zapier, instead of manually picking triggers and actions, I can just describe my workflow and Zapier figures it out.

Embedded Assistants

If AI Sparkles are like little touches of delight, Embedded Assistants are more like having a helper standing right next to you. Think of them as ChatGPT baked directly into a product, but with two differences: they can access your data, and they can actually do things in the product.

Two examples:

1. In Notion, I can ask questions and get answers from everything in my workspace — no need to search around.

   

2. In Shopify, there’s Sidekick. I can say “discount all my products 10%” or “How many orders did I get this week?” and it just does it or tells me.

   

AI Employees

Now, what’s next? Imagine that instead of just having an AI helper on the side, you could hire AI Employees inside your favorite product. You’d go to the “Teams” page in Notion, Linear, or Figma, click “Invite member,” and choose from a lineup of specialized agents. Each costs a monthly fee, each with its own specialty.

The distinguishing characteristic of AI Employees is that they are proactive. They also get the same access and permissions as a human employee, so they can start contributing right away.

For example, in Notion imagine being able to invite a “Product Manager” Agent for $75/mo. You can give it access to a page or the whole workspace. It’ll add comments with clarifying questions about workflows, turn complicated paragraphs into tables and diagrams, and do competitor research by browsing the Internet if it notices that the section is empty. If I have a specific ask, I can just tag it with a question or a request to do something for me.

In Figma, maybe you add a “Copywriter” Agent for $25/month. This one would specialize in writing better product and marketing copy, and ensuring it works with the designs already made. It might rewrite some text directly, leave comments with suggestions, or bring in examples from other products.

I don’t think we’ll see a singular super-intelligent AI that’s great for everyone and everything. More likely, some products will have a lineup of AI Employees, and whoever’s in charge will pick which ones to bring on board based on priorities, taste, budget, and other criteria.

Predictions

1. AI Sparkles will become standard and lose the “AI” label. They’ll just be part of what good software does.

2. Embedded Assistants will be embraced by companies with products that are complex — platforms, compound products, etc. The more complexity there is the more valuable an assistant is that can collapse complexity.

3. Pricing will change. Right now, you can add AI features for $5-10/month per user. That works for Sparkles and Assistants, but AI Employees will cost a lot more.² When it becomes clear that AI Employees can replace expensive human labor, it’ll be easier to justify premium prices.

1. This post isn’t about new AI-native products. It’s about how existing B2B products are already weaving AI into what they do — and how they might go even further in the future.

2. Charging an extra $5-$10/mo per user makes sense for AI Sparkles and Embedded Assistants because more employees = more usage = more incremental value. But AI Employees will be proactive, meaning they might spend a lot of compute researching a single question.