The new Powers of AI Assistants Explored
Real world examples from DoorDash, Pinterest, YouTube and more. Learn more about what they do, the technologies that underpin them - and how you can build your own.
🔒DoP Deep goes deeper into the real world experiences and features from top tech companies. If you’d like to upgrade to receive these in-depth pieces of analysis you can upgrade below. New reports are added every month.
DoorDash’s Chief Revenue Officer recently revealed that their new AI Assistant had a significant impact on their core metrics during testing: compared to traditional search, users who interacted with their Ask DoorDash Assistant built grocery baskets that were over 35% larger than typical grocery orders. Users also built grocery carts roughly 5x faster than doing so manually, often checking out in under two minutes.
Needless to say, AI Assistants are becoming increasingly powerful - and the product teams that use them effectively are delivering some pretty impressive results.
Not only that, but AI Assistants are becoming a new type of in-product search. Take, for example, some of the latest Assistants’ naming conventions: Ask DoorDash, Ask YouTube and Ask Pinterest are all new Assistants with the same prefix, signalling that they’re designed to improve the user experience by offering an additional paradigm for both finding things and getting stuff done.
In-product AI Assistants can now find information, answer questions and perform agentic actions on behalf of users.
It’s been well over six months since we last took a look at the powers of AI Assistants and in this Deep Dive we’ll dig deeper into recent examples of new AI Assistants shipped, exploring what new powers and capabilities they have, together with an analysis of the technologies that underpin them.
If you’ve shipped an AI Assistant in your own product and you’re curious about what new capabilities and features other companies have recently released, then this Deep Dive should help.
Coming up:
A full filterable gallery of all 35+ Assistants from companies like Instacart, Robinhood, Meta, Adobe and Notion, categorized by capability, so you can see exactly what’s out there before you build your own
The “ambient assistant” pattern showing up in SaaS companies, where AI starts working before you even ask
Why AI Assistants are moving out of the sidebar - and the cautionary tale of one product that got it wrong
How this analysis is structured
This analysis looks at over 35 different new AI Assistants across multiple different categories including B2C ecommerce, SaaS, finance, productivity and more.
Each example is categorized according to the new AI Assistant powers and capabilities with an explanation of how it works, together with a link learn more:
A filterable UI gallery of each Assistant
For each example, a filterable UI gallery is also included to help you to stay up to date with the latest UX / UI trends in product assistants:
The Assistant Powers explained
There are over 20 different types of AI Assistant powers and capabilities included. Here’s a snapshot of some of the most common across the 35 different examples:
Agentic Execution - rather than stopping at an answer, the assistant carries out multi-step actions on its own. For example, Robinhood now lets third-party AI agents execute real trades with guardrails in place, Replit Agent 4 builds entire application parts in parallel across isolated environments, and Meta’s Business Agent handles bookings for a business over WhatsApp without a human stepping in.
Knowledge Q&A - the assistant answers questions directly, whether that means general knowledge, live data, or a company’s own content. Siri can now pull answers from the web or your own messages and photos, Zendesk’s Copilots flag knowledge gaps for support teams, and Coinbase Advisor fields investment questions inside the app.
Memory and Context Awareness - the assistant carries information forward instead of making users repeat themselves. Airbnb’s support assistant loads a guest’s reservation details the moment a chat starts, Rippling AI resolves references like “my team” against real records, and Adobe’s Firefly studio has two dedicated memory systems, Elements and Projects, just to keep sessions and brand assets consistent over time.
Workflow and Task Automation - these agents run recurring tasks on schedules or triggers instead of waiting to be asked each time. Notion’s Custom Agents run 24/7 handling triage and status reports, AWS’s Kiro forces every feature through a spec-first workflow before code gets written, and Miro AI turns whiteboard diagrams straight into specs for coding tools like Cursor.
Shopping and Cart Building - turning a request into a ready-to-buy cart. Instacart’s Cart Assistant converts a photo of a handwritten list into a shoppable order, and Google’s Universal Cart lets you collect items from Search, Gmail, and YouTube into one basket.
Customer Support / Issue Resolution - handling service problems in place of a human rep. Zendesk’s Resolution Platform aims for full agentic resolution across messaging and voice, and Revolut’s AIR handles support questions under a zero data retention policy.
A closer look at each of the examples in more detail - with principles and takeaways for product teams to consider
Now let’s take a closer look at some of the examples in more detail, together with an analysis of the technologies that underpin them with some key considerations for product teams thinking about building their own or adding new capabilities to their in-product Assistants.