We’re building towards an “AI researcher” that research teams can rely on to execute qualitative research studies from end-to-end by gathering and analysing data from human participants. Central to this approach are agentic systems: AI-powered systems capable of independently performing tasks with varying degrees of autonomy.
Workflows vs. agents vs. multi-agents For some, "agents" means systems executing narrowly and predefined sequences of tasks. Others think of agents more broadly as highly autonomous systems capable of independently handling intricate tasks over time, constantly adapting the tools they use.
At Wondering, we consider both as forms of "agentic systems" but clearly distinguish two key categories: workflows and agents .
Agentic workflows use LLMs to execute tasks through predetermined sequences.Agents use some level of reasoning to autonomously determine the optimal approach to achieve their goals, dynamically select tools and methods, and then determine how to sequence these to achieve an outcome. Agents can excel at complex, open-ended tasks that are more complex in nature, as they can mix and match different workflows and tools depending on the task at hand. Today, these agents are often specialised on a specific set of open-ended tasks (such as completing literature review) to make sure they are performant.Multiple specialised agents can in some cases be combined to create multi-agent systems that are capable of performing an even wider set of open-ended tasks. It’s this multi-agent approach that we think is needed to build an “AI researcher” that can execute qualitative research studies end-to-end.
Start with agentic workflows To build the tools each agent in the AI researcher system needs to complete each step of a research study, we've launched agentic workflows across our product, including:
Our AI study builder workflows, which helps researchers define their research questions, and create purpose-built studies that collect the data they need. Our AI-moderation workflows, which are used in our studies to help researchers collect data from participants at scale through effective conversations and user tests. Our AI analysis workflows, which sift through all your study data to generate evidence-backed reports that directly answers your research questions. Let workflows to enable agents The above workflows automate significant portions of the research process already, and form the building blocks that can enable more powerful research agents. Next, we'll leverage these workflows to build research agents that, combined, can act as a multi-agent "AI researcher" and carry out research studies end-to-end by:
Independently designing and launching research studies based on high-level objectives. Autonomously recruiting participants and moderating conversations with participants without the need for guidance from a human researcher. Synthesising the data gathered, and delivering precise, evidence-backed research reports. That's the approach we take to build out agentic research tools that qualitative researchers can use to understand their customers and markets faster. We've started by building narrow agentic workflows and are optimising them so that they’re performant at specific tasks. We'll then build out capable agents can leverage these tools as they execute broader parts of the research process autonomously.
As these systems become increasingly capable, they will become the go-to tools researchers use to understand their participants. For this reason, we are sharing this approach to encourage early and continuous conversation in the research community about how best to use and refine these tools in the field of qualitative research.
.png)
.png)
.png)
.png)
