Compass
Where the Needle Points
Questions to ask yourself when you are building at the edge of AI.
This is not a roadmap, and not a list of products. It is the set of open questions we keep circling, each one big enough to hide a company inside. They pull on each other, and they will move as we learn.
✧Prologue · Why questions
Is the question the answer?
The questions on this page are not placeholders for answers. They are part of the work.
A sharp question travels further than the person who first asked it. Paul Erdős died in 1996, but he left behind a list of open problems, and researchers keep picking them up, working with a man they never met. Asking questions is a way to collaborate with future generations.
Exploration is not only about finding what is out there. Sometimes the most important thing you discover is a question, one that becomes the end of one journey and the beginning of another.
This Compass is how we make those questions public: the ones pulling us forward right now. Vivid questions are keys opening unexpected paths.
01Agent-first apps
When agents do the work, what apps do they need?
Most software exists because humans do the work: Gmail for communication, calendars for planning, Premiere Pro for video editing. As agents take over that work, which parts of this app layer remain useful, which disappear, and what has to be invented?
When the main user is an agent, what replaces the interface: an API, a protocol, a conversation? And what does good UX even mean for a user that reads the documentation in milliseconds and never gets lost?
02Proactive AI
What would it take for AI to act before we ask?
Almost everything AI does today begins with a human prompt. What signals would let it recognize what is coming early enough to prepare, without waiting to be asked?
Where is the line between an agent that anticipates and one that interrupts? What would it take to trust an AI that starts moving before you ask?
03Temporal context
How does an agent learn what time means?
Agents wake up knowing nothing about deadlines, while the human world runs on payroll cycles, quarterly objectives, day and night, people asleep in the loop, and logistics that can never be instantaneous. How should an agent represent those rhythms?
All of an agent's memory points backward; what would it mean to remember the future? Could an agent feel the difference between a deadline tomorrow and one next quarter?
04Post-AGI compute
Where does the work actually run?
Intelligence may live in the data centers of a few big labs, but actions still touch a CRM, a command line, or a file on a machine somewhere. Should that execution live on the employee's computer, in a cloud, or on racks inside the company?
Will the pull of privacy and cost drag agent work back on-prem, or will everything melt into a few clouds? And what happens to the employee's computer when the employee is an agent?
✦The answers
We do not just stare into these questions. We build answers.
We do not claim to have solved them. We build against them, put our hypotheses in people's hands, and let the work sharpen the questions.
AgentCal is our first attempt: a calendar made for an AI agent, so it can hold deadlines, coordinate milestones, and work toward what is coming.
Your AI's calendar. So it can plan ahead.
youragentcal.com
AgentCal · Your AI's calendar
Give your AI agent a real calendar of its own: it plans deadlines and milestones, works on schedule, and you follow along live.
Every answer sharpens the questions. If you are chasing one of these, or you already hold an answer we should read, come talk to us.