As Hootsuite explored a new generation of AI agents and automations, Product needed to understand what customers actually wanted from them. I led a mixed-methods study on how users adopt AI today, what they expect agents and co-pilots to mean, which tasks they trust AI to run on its own, and how their usage might grow over time. I interviewed 10 customers and triangulated their feedback with customer feedback data, industry research, a competitor review, and product usage data, then translated the findings into recommendations for Product and Design.
Hootsuite was considering what kinds of AI automations to build and how its AI experience should be structured, including whether to ship a single agent, several specialized agents, or a co-pilot. These decisions were being made while customer workflows were changing quickly, and there was little shared understanding of how users thought about AI, what they trusted it to do, or what the product’s AI terminology meant to them.
The research focused on one primary question:
Which tasks do users trust AI to run on its own versus assist with, and how should AI agents be named, organized, and supported to fit the way customers actually work?
The study also covered current AI adoption and confidence, the vocabulary users use for agents, co-pilots, models, tools, skills, and modes, appetite for a single AI-first surface, and how users understand AI consumption.
I partnered with Product and Design to turn open strategic questions into five research goals: current-state AI adoption, task fit and trust in automation, taxonomy perception, appetite for AI-first workflows, and the value of single versus multiple agents. Because the topic was new and fast-moving, I paired in-depth interviews with secondary sources so interview themes could be checked against what customers say and do at scale.
Defined five research goals and a 60-minute interview guide with Product and Design, covering workflows, trust, terminology, and agent structure.
Spoke with 10 sales-enabled and self-serve customers across industries and Hootsuite products from July to August 2026.
Compared interview themes with customer feedback, industry reports, a competitor review, and product usage data.
Consolidated findings into 9 recommendations and delivered a report to Product and Design.
Integrate best-in-class models rather than compete on quality, and differentiate on grounded answers, citations, and data governance.
Rename the taxonomy so labels read as capabilities, and design agents around the role users expect.
“It’s not about trusting AI… it’s about trusting the people using it.”Enterprise participant
Offer per-task autonomy (suggest, act with approval, act and notify), default to approval, and always provide an easy undo.
Repeatable, low-stakes, reversible
Judgment, facts, money, relationships
Prioritize memory that learns from corrections and give enterprises a security-first story: closed models, no data training, and admin controls.
Make consumption transparent and add prompt aids that show users what is possible.
Consumption meter & estimates
Prompt aidsThe report turned a broad set of open questions into 9 recommendations, each with an owner and impact rating and mapped to the product strategy pillars. Six were rated high impact, including per-task autonomy controls, trust signals such as citations and reasoning, governance controls for admins, and memory that learns from corrections. The research gave Product and Design a shared vocabulary, an evidence-backed view of where users will and will not delegate to AI, and a clearer basis for deciding when to ship an agent versus a co-pilot.
