Introducing Offloop
Jin, Fred and Yum
AI can now do more of the work than ever before.
It can research a market, write a campaign, analyze a customer account, build software, create a financial model, and operate across the tools a business depends on.
But running a business with AI still feels strangely manual.
You decide what should happen next. You open a chat, explain the context, assign a task, review the answer, move the result somewhere else, and turn it into the next prompt.
When the work is waiting on a customer, a deadline, or a decision, you have to remember to come back.
The AI may do the task.
But you still run the loop.
As agents become more capable, this becomes the real limitation. The problem is no longer only whether an agent can complete a task. It is whether an agent team can take responsibility for an outcome, keep working as circumstances change, and involve you only when your judgment is required.
A new kind of company
We believe the next great company may be built by a small team working with a team of agents.
Not assistants waiting for one prompt at a time.
A real team: agents with different capabilities and responsibilities, working across the business, responding to what changes, coordinating with one another, and continuing to make progress without requiring a person to initiate every step.
The human team will define the ambition, establish the principles, and make the decisions that require judgment.
The agent team will take responsibility for the work between those decisions.
A customer shows interest. The team follows the opportunity from first response to close.
A launch needs creators. The team researches the market, runs outreach, negotiates within budget, manages delivery, and improves the campaign.
Customer requests begin to form a pattern. The team identifies the opportunity, gathers the evidence, prepares a proposal, and brings a tested decision back to the product team.
You do not tell the agents every step to take.
You define the outcome and the boundaries.
The team keeps the work moving.
This changes more than productivity. It changes how much a small team can accomplish, how quickly an idea can become an organization, and who can build a company in the first place.
The missing environment
Most AI products begin with a prompt and end with an answer.
Real businesses do not work that way.
Work begins with goals and signals. It moves through people, tools, decisions, dependencies, and time. It pauses while waiting for the outside world. It changes as new information arrives. It ends only when the outcome is reached.
Chats are built around conversations. Agent runtimes are built around individual runs. Project tools still assume that people will carry the context, coordinate the work, and remember what happens next.
None of them gives an agent team a durable place to take responsibility for an outcome over time.
A chat can remember what was said without knowing what the business accepted. A run can record what an agent did, then end before the surrounding work is complete. A project tool can track tasks while still relying on a person to create the plan, chase dependencies, and initiate every next step.
The missing piece is not another agent that can complete another task.
It is an environment in which people and agents can operate as one continuing team.
A place where the goal survives the conversation. Where decisions, evidence, ownership, and next steps remain attached to the work. Where waiting work returns when the world changes. Where agents can act independently within clear boundaries and know when to bring a decision back to a person.
That is what we are building with Offloop.
The workspace for the whole loop
Offloop is a collaborative AI workspace where a small team can hand business outcomes to an agent team.
You describe what you want to happen and define the boundaries that matter.
Offloop keeps the loop moving.
Your agent team turns the outcome into work, brings in the right capabilities, operates across your tools, follows up on what is waiting, and adapts as circumstances change.
When the team can continue independently, it does.
When it needs your judgment, it brings you the decision, the relevant context, and the evidence behind its recommendation.
You make the call.
The team continues.
You do not have to remember to restart the work. You do not have to reconstruct the context. You do not have to turn every completed task into the next prompt.
Other AI helps you complete tasks.
Offloop gives an agent team responsibility for carrying the business forward.
From signal to outcome
Consider a new customer lead.
In Offloop, the lead can become a scheduled meeting, a discovered product requirement, a scoped engineering project, a tested integration, a signed agreement, and a closed opportunity. The agent team carries the context across the entire customer journey. A person joins when attendance, commitment, or signature requires human authority.
Or consider a creator campaign.
A small team defines the target, timeline, budget, and exception policy once. Agents research the market, run outreach, compare real offers, negotiate within their authority, manage delivery, and monitor results. Only terms outside the agreed boundaries return for a human decision.
Or consider a product opportunity.
Signals arrive independently from support, sales, and the public internet. An agent team recognizes the common need, gathers evidence, defines the first version, prepares a prototype, coordinates implementation and QA, and brings back a tested staging decision.
These are different kinds of work, but the loop is the same:
- A goal or signal enters the workspace.
- An agent takes responsibility for the outcome.
- The team divides and executes the work.
- Waiting work resumes when something changes.
- Decisions return to the people with the authority to make them.
- The team verifies the result and closes the loop.
The user does not become the project manager for a collection of agents.
The agent team owns the work between decisions.
Built for agent teams
We are building Offloop around four systems designed for continuous agent work.
Agent Interface gives agents a shared way to participate in work, use tools, report progress, return evidence, and ask for decisions.
Dispatcher Model turns an outcome into coordinated work, routes it to the right agents, and adjusts the plan as new information arrives.
Identity Awareness Harness preserves who is acting, what each participant owns, which context applies, and where their authority ends.
Persistent Cloud Runtime allows work to keep running, waiting, and resuming after the original request or session has ended.
Together, these systems let Offloop organize around the lifecycle of the work, not the lifetime of a single agent run.
This distinction matters. More context is not automatically better context. More agents do not automatically make a team. More autonomy is not automatically useful.
An agent team needs the right context, clear responsibility, complementary capabilities, visible evidence, and explicit authority boundaries. It needs to know when to continue and when to ask.
Does an agent team produce better work?
We also want to know whether structuring work across an agent team can improve the result itself, not only make the process easier to follow.
We have started evaluating early Offloop configurations on benchmarks designed around professional work, including JobBench, GDP.pdf, and GDPval-AA.
These evaluations test whether complementary responsibilities, inspectable intermediate work, and evidence returned into shared context can improve what the team produces.
A benchmark still begins with a prepared task and a defined finish line. A real business does not. Goals change. New information arrives. Permissions and priorities move. Work pauses and returns.
The larger test is whether people and agents can remain coordinated through those changes.
That is the test Offloop is being built for.
Join the waitlist
Offloop is still early.
Today, we are sharing what we are building and opening the waitlist to the first small teams that want to hand complete business loops to agents, not just ask AI for individual tasks.
There is much more to build. Agent teams need better ways to learn how a business operates, develop complementary capabilities, coordinate over longer periods, and understand when to act and when to ask.
People need better ways to grant authority, redirect work, and understand the state of a business without managing every step.
But we believe the direction is clear.
The future will not be built by people prompting agents one task at a time.
It will be built by small teams that can give an agent team responsibility for an outcome and trust it to keep the work moving.
Offloop is where that team works.