Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
AI of the now: Designing for Agents
Summary
AI agents are on the rise: people are talking about them, getting excited, and putting infrastructure into place to support them. You may be asked to design agents soon—or you may want to think through your product as an agent in order to get a leg up on your competition. What’s different about designing for agents? Are you ready? Rosenfeld author Christopher Noessel published Designing Agentive Technology back in 2017, and it seems like the world is just now catching up. Join Chris as he recaps the core ideas from that book, how agents have evolved since publication, and what it all might mean since the explosion of generative AI. Takeaways include… What is an agent? How do agents differ and complement other modes of interaction? Why are people getting excited about it now? What does generative AI bring to the agentive table?
Key Insights
-
•
Agents act on behalf of principals, whether humans, animals, or machines, performing tasks with variable human attention involvement.
-
•
Agentive technology operates largely outside the user's attention, unlike assistants that require active engagement.
-
•
Chris identified 27 unique agent use cases including setup, monitoring, handoff, and disengagement.
-
•
There are four modes of agency: manual, assistant, automation, and agentive, depending on who performs the work.
-
•
Agentive systems don’t require an agentic backend, and agentic backends can exist without agentive front-ends.
-
•
Recent AI advances use an 'agentic backend' architecture where a large language model orchestrates specialized agents for complex tasks.
-
•
Design challenges include enabling user control over agent plans, forbidding specific data sources, and handling persistent queries.
-
•
AI risks include deskilling users, overreliance on AI output, and ethical concerns from hidden, biased, or commercialized agentic actions.
-
•
Agents hold huge value potential by performing work outside users’ limited attention, making them a strong business opportunity.
-
•
Despite a rise in AI backend engineering, design is essential to maintain usefulness and address human-AI interaction complexities.
Notable Quotes
"An agent is anything that displays agency, the ability to take actions and choose which actions to take."
"I don’t want a surgeon to tap me on the shoulder and ask where to put the incision; that’s their expertise."
"My Roomba vacuums while I’m at work. It does its work out of my attention."
"We’ve been dealing with assistance since spellcheckers in the 1970s evolved to grammar checkers and beyond."
"The user might want to see that plan in advance and make sure they’re okay with how it’s going to go about it."
"Large language models sometimes hallucinate, so agentic backends assign sub-tasks to specialized expert agents."
"Agents do work outside the user's attention, freeing up their limited time for other activities."
"We risk deskilling users and overreliance, where people rubber stamp AI output without critical thought."
"Agentic systems could be hidden, allowing corporations to push self-interested biases undetected."
"AI should augment humans, not replace them, especially given societal disparities between deep and fast thinkers."
Or choose a question:
More Videos
"We wanted them to see the government not as an old machine, but as people who see themselves as users."
Sofía Delsordo Kassim VeraPublic Policy for Jalisco's Designers to Make Design Matter
December 8, 2021
"If it’s not usable it’s not valuable."
Product and Design at Bloomberg: A 15-year Evolution
December 6, 2022
"Unmoderated research is exciting because it lets us gather insights while freeing people up to be in two places at once."
Liza Pemstein Jane DavisScaling Research Via an Ops First Model at Clever
March 27, 2023
"Openly sharing stories helps us learn from those who came before us and carve new paths."
Kara KaneTheme One Intro
November 16, 2022
"Many organizations silo data inputs and prioritize within categories before combining results."
Mark Interrante Harry MaxAI for Prioritization (3rd of 3 seminars)
July 11, 2024
"A good two-factor system offers multiple methods, like authenticator apps, calls, and texts, with generous timeouts."
Sam ProulxOnline Shopping: Designing an Accessible Experience
October 3, 2023
"Product managers are going to have to say no to roughly 95% of everything that comes in."
Rich MironovHow Can Product Managers and UXers Help Each Other (and Why are Product Folks so Annoying Sometimes)?
December 6, 2022
"If someone says hell no, don’t stop, keep going."
Juhan SoninDesign Now! The Agenda for Action
September 4, 2025
"Safety is a fundamental human need which is often taken for granted and maybe that’s why it’s forgotten about."
Mansi GuptaDrawing from Feminist Practice to Make Inclusive Design Operational
September 9, 2022
