Models
Reading time:
2
min

Software development now happens at a pace that was previously unimaginable. Agents can do the heavy-lifting for the vast majority of projects. What once took months now takes hours, and what once required a large team of experienced engineers can now be done by a single person.¹
In this new world, the scarce quantity is no longer code, but rather human attention. Real value is created only by tightly aligning what you build with the needs and preferences of your users. To achieve such alignment, you need people to use your product in a serious way, so that you can study how they use it, and you need them to provide considered feedback on their experiences. However, this is incompatible with the rapid pace of development. Designing and running user studies is slow and expensive. People are busy and distracted. Your product is changing so fast that any feedback you get is irrelevant by the time it arrives.
Building software has never been easier. Building software that users love has arguably grown more difficult. Traditional user research can’t keep up with the pace of development, and superintelligent testing agents mislead us more often than they inform us.
Can agents serve as a proxy for human attention? Agents can find bugs and security vulnerabilities in software, and so perhaps they can also provide rich feedback on the experience itself – missing functionality that users will expect, design antipatterns, information overload, and so forth.
Unfortunately, this is where the superhuman capabilities of today’s agents become a liability. The testing agents one gets from frontier providers like OpenAI and Anthropic do not explore interfaces or consume information the way humans do, and so their behaviors and reactions are often completely disconnected from those of human users. For example, whereas humans selectively attend to parts of a webpage, superhuman agents consume the entire page and remember it perfectly. These agents never overlook instructions, try to click on inactive page elements, or grow bored of typing. If your chat app produces a wall of texts, superhuman agents will read it from start to finish; human users will tend to skim, at best.
Thus, the net effect of agent-based testing is often a distorted picture of how human users will respond. You charted out a happy path through your experience. Your superhuman agents followed it perfectly and loved what you built. You deployed it to your users, and they were lost, frustrated, and unsuccessful. You might never win them back.
For example, whereas humans selectively attend to parts of a webpage, superhuman agents consume the entire page and remember it perfectly. These agents never overlook instructions, try to click on inactive page elements, or grow bored of typing.
Bigspin saves you from this trap. Bigspin’s Poser agents act and react the way humans do, because they are trained on real human data and informed by extensive research on human preferences, goals, and behavioral patterns. Superhuman testing agents might sail through your experience, express delight at your chatbot’s detailed messages, and provide exactly the right information at the right time. Poser agents will give you a picture of the actual human experience: what people will notice, skip, and miss entirely; where they will feel friction and confusion, and where they will be delighted.
Importantly, Bigspin’s Poser agents can operate at the speed of AI development:
When placed in the inner-loop of your own development process, Posers will provide feedback directly to your coding agents about the interfaces and functionality they are creating. With help from Posers, your coding agents will produce software that works for humans right from the start. Bigspin will provide a rich report to you on the guiding principles behind its design and the antipatterns you avoided along the way, and it will show you full video replays of agents using your product.
On Bigspin, you can quickly design and run Poser-based studies of your product, website, or prototype. Such studies are ideal for evaluating proposed design changes and weeding out problematic antipatterns. They provide all the affordances of human studies, including detailed feedback from individual participants and full video replay of their usage, but they are completed in minutes, and at a tiny fraction of the cost of human testing. Think of all the user testing you would conduct if you had endless time and resources – those can all be run easily with Bigspin.
Bigspin can complement automatic testing using Posers with traditional user testing. We remove all the friction of setting up human studies, running them, and analyzing the results. In addition, each human test you run has higher value than before, since the lessons are fed back to your Poser agents so that they know more about your goals and experiences.
All Poser agents are highly customizable, so you can conduct tests involving groups of simulated participants with different goals, backgrounds, and effort dynamics. Panels of agents with different characteristics can be created automatically with Bigspin, to meet your specific research objectives.
Building software has never been easier. Building software that users love has arguably grown more difficult. Traditional user research can’t keep up with the pace of development, and superintelligent testing agents mislead us more often than they inform us. With Bigspin, you’ll catch the problems you’d otherwise learn the hard way, before you launch.
Getting started couldn’t be easier:
Add your prototype or live URL, add your research goals, customize your personas.
Watch Posers navigate, narrate, and interact with your product.
Get actionable insights, video clips, and recommended opportunities.
Start shipping products users love right from the start.
¹ Or, more precisely, their agents

