Member of Technical Staff

About Blockit

We are a Sequoia-backed startup building the AI agent for the world's time.

Time is the most valuable resource we have. The calendar is the database for this resource -- yet it is 1/ dumb (missing core database primitives -- typing, indexes, clustering, etc) and 2/ it is disconnected (non-interoperable with other calendars).

Sadly, calendars are based on a ~50 year old email protocol (SMTP), which has 4.8 billion users(!), but literally no one owns it, and thus no one can change it. Until Blockit :)

We are using an agent to make calendars 1/ intelligent and 2/ interconnected . LLMs unlock (1), which then paves the way for (2). Think of this a little bit like Slack Connect for time, where a calendar network would allow companies to automatically coordinate calendars externally .

Blockit is a fully autonomous agent that runs around email threads and slack channels to handle the end-to-end complexity of scheduling -- timezones, group coordination, in-person logistics, etc.

Sometimes coordinating a very complex meeting across many external, busy people can take days, seriously! Blockit collapses that to minutes. And as more people connect their calendars, our agent-network becomes exponentially more powerful.

Technical Challenges

Engineers sometimes think of scheduling as a search problem -- but it's not. It's a constraint satisfaction problem. Our customers calendars are 120% full and balancing multiple unstated invariants and constraints. Meetings have to move in order to find space for other meetings based on priority.

While every LLM application to date has been a one-on-one conversation, Blockit is one of the first multiplayer , stateful AI agents -- it has to simultaneously keep track of numerous people coordinating with it across channels. It is effectively coordination game theory, while preserving calendar privacy across companies.

The context management challenges are awful. Our problems are right at the frontier of multiplayer agent orchestration , which OAI/Anth haven't figured out how to solve yet. 

Links

This is the foundation of a platform of AI agents with access to the world's time. We've raised significant funding from Sequoia , and we're a small, sharp team that moves fast, ships constantly, and holds a high bar. If you want to build something genuinely new, we'd love to talk.

See the following links to learn more about us:

• Techcrunch -
• Sequoia Thesis -
• Teams page -

The role

You’ll own the intelligence at the core of Blockit: our scheduling agents that autonomously coordinate meetings across people, time zones, and constraints.

This includes designing and iterating on agent architectures, writing and refining prompts, building evaluation frameworks, and shipping new capabilities as models improve. You’ll work across our orchestrator agents (which manage conversation flow) and specialized sub-agents, with the goal of making Blockit smarter, faster, and capable of handling increasingly complex coordination problems.

You’ll be architecting and building real-world AI agents used in production.

What you’ll do

• Write and refine prompts across our agent system—orchestrators, sub-agents, and tools
• Build and maintain evals to measure agent quality and catch regressions
• Debug agent failures: figure out why it misunderstood a request or made a bad call
• Implement new agent capabilities as user needs expand
• Experiment with new architectures and techniques as models improve
• Instrument and analyze agent behavior to find patterns and failure modes

What we’re looking for

• 2+ years of experience shipping and owning production software
• Strong backend engineering skills, with the ability to work across the stack when needed
• Experience working with LLMs in production systems (or a demonstrated ability to learn quickly in this space)
• Deep curiosity about agent architectures and how the industry is evolving beyond simple prompt-based systems
• Clear, structured communicator who can explain what’s working, what isn’t, and why

Location

• San Francisco, CA. On‑site 4 days per week (WFH Wednesday's)