Know the job
The task has a purpose, a scope and a place in the wider project. The agent gets the context needed for that work.
Workflow OS / People, agents & the work between them
The Future Is Already Here!
Workflow OS connects human direction with AI-assisted delivery. The brief, the working context, the tools and the evidence stay connected as an idea becomes a working result.
An illustration of the delivery flow, not a live feed or a claim that any model guarantees a correct result. Tool access, checks and human decisions govern each task.
01 / Why we built it
Keep the context.
Move the work forward.
Running several products creates work between the work: explaining decisions, finding the right context, coordinating changes and checking what actually happened.
Adding AI makes that coordination more important. An agent can act quickly on an incomplete instruction. Someone still has to connect the business goal, the current system and the result that matters.
We built Workflow OS to make our delivery method explicit and usable by both people and agents. It brings strategy, operating rules, task flow and visibility together around the products we run.
02 / People & agents
A defined role.
A person to answer to.
Our “digital employee” approach gives an agent a job, a bounded set of tools and a human owner. George coordinates delivery work within this operating system.
Work inside boundaries
Permissions, operating rules and execution controls define what the agent can do and when it needs to stop.
Keep a human owner
People set direction, resolve trade-offs and review whether the delivered experience meets the intended need.
Automated where it should. Human where it matters.
03 / What we built
A system around
the whole job.
Workflow OS connects the human work surface, the engineering context, the execution system and the evidence used to review progress.
- 01
Direction & decisions
ClickUp holds the brief, priorities, ownership and human decisions.
- 02
Shared context
Lifecycle documentation and GitHub connect operating rules, product knowledge and implementation.
- 03
Coordinated execution
Cloudflare services coordinate eligible work with bounded agent runs and GitHub build and release checks.
- 04
Visible evidence
The Command Center connects task sessions, recorded activity, release evidence and work waiting for people.
The Command Center is an internal operating surface. This page illustrates the approach without exposing private tasks, credentials or customer information.
04 / Research & development
Making agent work
dependable enough to use.
01
How does an agent know what is true now?
What we develop and test
Bring the live task, repository, operating rules and relevant decisions into the work. Separate historic reference material from the current state of the system.
What we learn from operating it
Context needs an owner and a source. A confident answer based on an old brief can still produce the wrong change.
02
How do several agents work without colliding?
What we develop and test
Route work to a defined repository and account for dependencies, capacity, costs and activity already under way. Keep the identity of each execution attempt with its evidence.
What we learn from operating it
More simultaneous activity does not necessarily produce more useful progress. Work needs coordination, clear boundaries and a way to pause when conditions change.
03
How do we know the work actually succeeded?
What we develop and test
Connect the change, build checks, deployment and live read-back to the task. Present the result to the person responsible for judging whether it meets the need.
What we learn from operating it
A completed command and an accepted outcome are different milestones. Reliable delivery needs evidence from the system that was changed.
05 / In use, still developing
Our own work.
A practical proving ground.
We use Workflow OS within the Group’s product and engineering operations. Operating it ourselves exposes the failures, handovers and ambiguous decisions that guide the next iteration.
Current development focuses on execution boundaries, supervision, reliable evidence and the cost of useful work. The system continues to evolve as models, tools and the products it supports change.
Connecting operation with assurance
Workflow OS provides the working context for agents. Trust Center’s research explores the evidence and controls needed to assess particular agents in particular situations. Their connection is a development direction, with evaluation still to do.
Our track record
Experience earned
by doing the work.
Client delivery. Products of our own. The everyday work of making them run. Each gives us a different perspective on what makes technology useful.
Explore our delivery track recordGrounded in real businesses
Our team’s work connects brands, creators and educators with their audiences, through digital platforms and the systems behind them.
Built and operated by us
We live with the products we create. Supporting them beyond launch informs the decisions we make on the next one.
Turning experience into what’s next
Recurring problems become reusable technology. The questions still unanswered become the starting point for our research.