A product thesis, with evidence and gates
A world worth opening.
A product worth paying for.
Help operators understand their agent teams and explain the work to clients. The visual world is the interface. Trustworthy observations, useful decisions and clear deliverables must earn the subscription.
LOCAL PROTOTYPE DEMAND & PRICING UNVALIDATED
No customers, revenue, market size or probability of success claimed. Research identifies substitutes and technical options; real users and paid renewals must establish commercial viability.
The first wedge to test
Who
Small automation agencies
Operators managing several client workflows. Test whether explaining exceptions and handoffs is frequent, expensive work.
What
Client-safe mission rooms
A shared understanding of what ran, what was reported, what needs a decision and what was delivered. Client publishing and approval features remain planned.
Why return
Less coordination work
Measure time-to-find a blocker and decision accuracy against a plain table and existing tools. Delight supports utility; it does not prove it.
What makes the product stronger
Working foundation
Original customizable world, persistent team import, authenticated event reporting, full deliverables, native background execution, and a real four-node LangGraph team with parallel work. Users can now pair a runtime, launch registered tasks, inspect run history, and request acknowledged controls at step boundaries.
Opt-in exports. Execution stays in its own runtime. Runtime claims remain distinct from independent verification.
Next defensible workflow
Exception → evidence → responsible operator → explicit decision → reviewed handoff.
Test demand for selective client views, permissioned approvals, branded reports and mixed-runtime connections. These are product hypotheses, not unique features or a moat already won.
Competition changes the strategy
| Category | Examples | Our implication |
|---|
| Visual agent worlds | Pixel Agents, Star Office UI, Agents Office | Characters, rooms and customization already exist. Original art is necessary, not a moat. |
| Observability | LangSmith, Langfuse, AgentOps | Integrate with mature engineering tools. Win a more accessible operator/client workflow. |
| Orchestration | LangGraph, CrewAI | Preserve existing runtimes and control semantics. Avoid demanding a migration. |
Commercial competitor entry offers include LangSmith Plus $39/seat/month plus usage, Langfuse Core $29/month, and AgentOps Pro starting at $40/month. Billing units differ. These offers do not establish willingness to pay for Forge.
The next 90 days, earned in stages
01 / UsefulDays 1–14 · real teams and comprehension
Three external teams connected. Median setup under 15 minutes. Four of five observed users identify working, failed and disconnected states correctly. Interview ten operators about actual recent incidents.
Days 15–30 · narrowly scoped design partners
Test a read-only client brief. Seek actual payment from two of three partners and weekly use for four weeks. Add an adapter when pilot users need it, not to grow a logo wall.
Days 31–60 · identity, isolation and reliable evidence
Membership authorization, selective read models, token revocation, quotas, retention, backups and an isolation review. Controls appear only when the runtime acknowledges them.
Days 61–90 · activation and renewal
Onboarding without founder help. Measure support time, useful weekly activity, contribution margin and renewal reasons. Reposition if people only come for novelty.
Where money could come from
Price hypothesis
Builder
$29 / monthHistory, useful alerts and saved worlds. A free local edition can support adoption.
Price hypothesis
Studio
$79 / monthMultiple worlds, team collaboration and longer history after access controls exist.
Price hypothesis
Agency
$199 / monthBranded client views, approvals and handoff reports. These features are planned.
First test $300–$1,000 setup engagements with tight scope and recorded labor. Keep inference on the customer's provider account. Cosmetics can be secondary revenue. No checkout, unlimited promise, automatic overages or SLA offered.
Explore the economics
Illustrative assumptions, not a forecast or infrastructure quote. Support includes founder labor at $40/hour. Development, acquisition, overhead, tax, churn and refunds are excluded.
Hosting without risking the rest of KU13SH
Showcase
www.ku13sh.com/agent-forge/
A proposed static marketing or read-only showcase. The current site uses a static S3/Amplify release. No existing site files were modified or published in this pass.
Future authenticated app
forge.ku13sh.com
A separate origin and backend for the API, event store and workers. External runtimes connect outbound over HTTPS. Never host a shared service on the local Codex login.
Public beta needs membership-based authorization and isolation, explicit retention and secret filtering, restore-tested backups and a hosting budget. The static host alone does not run this Python stack.
What actual user feedback changes
A small firsthand sample, not a representative survey: the Pixel Agents VS Code review tab contained enthusiasm for enjoyment and awareness, a request for clearer status, and a historical configuration-loss report. The maintainer stated that configuration issue was fixed in v1.4.1; we do not treat it as a current defect.
Selected issue reports also describe stale presence, wrong character-to-session mapping and crowded labels. Forge's response is explicit pairing without settings edits, stable run identities, acknowledged controls, readable inspection and connection status separate from task completion.
Read the original reviews ↗ · Evidence, limitations and acceptance checks ↓
The strongest path is a sequence of honest tests: useful to real operators, safe with real data, paid for repeatedly. If those gates fail, change the product before increasing spend.