Delos vs win.sh: Two Operating Models
Delos centers role-based AI workers with professional identities. win.sh runs a 24/7 business loop inside an authority matrix.
Founder, Task Machine
Recurring agent work needs more than a model that can complete a task. It needs a durable owner, boundaries, evidence, and a review rhythm that still works when the operator is busy. The central choice is whether to manage a goal assigned to a specialized worker or a continuous loop that watches the business and decides what to do next.
Delos centers a catalog of role-based AI workers with professional identities and business-tool access. win.sh centers a 24/7 monitoring and action loop governed by a per-work-type authority matrix. Both can produce useful agent work, but they ask the operator to shape and review it differently.
How do Delos and win.sh compare?
| Dimension | Delos | win.sh |
|---|---|---|
| Core abstraction | A goal assigned to a role-based AI worker | Continuous business monitoring and action loop |
| Human role | Set goals, connect tools, follow activity, and handle escalations | Set rules, review decisions and receipts, and adjust authority by work type |
| Execution | Hosted workers with professional identities and access to Slack, Teams, and more than 3,000 tools | Cloud sandboxes connected to accounts the customer owns |
| Strongest fit | A specialized role that benefits from its own email, phone, memory, and tool access | Low-touch operation with a compact morning review rhythm |
| Recurring work | A worker runs goals continuously and retains role-specific context | A loop runs continuously, with recurring workflows and a daily brief |
| Control | Goals, access boundaries, activity, and escalation | Approval gates, authority matrix, budget cap, receipts, and Decisions tab |
| Main cost | A process crossing several roles or gates can be less explicit | The loop acts before the brief, so some control happens after execution |
| Commercial model | Team plan at €50 per month for 40,000 credits, with tailored Enterprise pricing | Self-serve monthly budget from $50 to $10,000, no revenue share |
Who does Delos fit?
Delos fits companies that want one specialized AI worker to own a recognizable business role. Its catalog, fast setup, professional identities, memory, and built-in channel presence reduce the work required to put that worker into existing tools.
The limitation follows from the same choice. A goal and activity view can be enough for a contained role, but a process crossing departments, artifacts, or approval boundaries may need a more explicit shared model. Test how missing input, failed checks, and handoffs appear before choosing the worker as the primary object.
Who does win.sh fit?
win.sh fits founders who want connected business systems watched continuously. The loop proposes actions, proceeds inside standing authority, holds risky categories for approval, and compresses activity into a morning Telegram brief and Decisions tab. Approvals, edits, and rejections become rules, which lets trust grow separately for each kind of work.
The ownership model is a strength. win.sh connects to accounts the customer owns, applies a hard monthly budget cap, shows dollar-based receipts, and takes no revenue share. Its tradeoff is the default rhythm: the loop runs before someone opens the product, and review often follows execution rather than steering each step in flight.
Where does each product put human control?
Delos puts control around a specialized worker through goals, connected tools, activity, and escalation. win.sh moves more control into standing policy. The authority matrix describes what may happen, approval gates stop selected risks, and the brief summarizes the resulting work.
Neither approach is universally safer. A carefully configured specialist worker can be easier to bound than a general loop. A mature authority matrix can be easier to operate than dozens of hand-built review branches. Compare the actual exception paths, not the autonomy label.
Where does Task Machine fit?
Task Machine takes a third position for operators who want agents to execute while the recurring process stays explicit. Chat sets direction, one inbox gathers approvals, questions, failed verifiers, proposals, and exceptions, and tasks preserve detailed state. Workflow graphs record branches, human gates, verifier gates, and append-only step history.
That model involves the operator more than win.sh and requires more setup than Delos when one role already fits the job. It does not replace Delos workers with their own phone and professional identity, and it does not promise a low-touch morning-brief rhythm. It fits when work crosses humans and agents and the process must remain readable independently of who executes it.
Which should you choose?
Choose Delos when a specialized AI worker with its own channels and memory should own the goal. Choose win.sh when continuous monitoring, accounts you own, a hard spending cap, and a light daily review rhythm matter most. Choose Task Machine when explicit workflows, verifier-backed gates, and one inbox for every judgment call matter more than minimum operator involvement.