Keine Idee aus den heutigen Storys hat die Messlatte geschafft. Eine schwache Idee zu veröffentlichen ist schlechter als keine.
Gesichtete Storys
25
Gründlich gelesen
7
Vorgeschlagene Ideen
14
Übrig geblieben
0/14
Keine Idee hat die 6.0-Linie überschritten.
Verworfen · 14 · unter der 6.0-Linie
| 5.9 | Pay transparency compliance checker for job ads | Scored 5.9, under the 6.0 bar. |
| 5.6 | Active exploitability verification for new CVEs | Scored 5.6, under the 6.0 bar. |
| 5.4 | Differential fuzzing service for auth and libc behavior | The buyer set (Alpine-derived image maintainers, OpenWrt vendors) is a handful of organizations, OSS-Fuzz is free, and the work is consultancy-shaped — nowhere near a venture-scale market. |
| 4.9 | Interactive MCP training for engineers and managers | Too weak on moat (3/10). |
| 4.9 | MCP auth and permission gateway for small teams | Near-duplicate of the recently published unified permission and audit layer for agent access, and RunLayer/WorkOS are already funded and shipping MCP auth gateways. |
| 4.6 | Scanner for data-exfiltration patterns in agent skills and MCP servers | Invariant Labs' open-source mcp-scan already scans MCP configs and skills for exactly these prompt-injection/exfiltration patterns, and funded AI-security vendors (Lasso, HiddenLayer, Cisco) are expanding into the same surface. |
| 4.4 | Sandboxed runtime with egress control for coding agents | Agent harnesses are baking in native sandboxing (Claude Code's sandbox mode) while E2B/Daytona own cloud sandboxes and gVisor/devcontainers are free, so a cross-platform third-party runtime gets squeezed from both sides. |
| 4.3 | Total-comp job ad database for UK and EU tech | levels.fyi and Glassdoor already occupy this space, the very disclosure mandates cited as 'why now' put the data into every ad for free (shrinking the gap), and job-hunting candidates notoriously don't pay subscriptions. |
| 4.3 | Egress guard that blocks PII leaks from coding agents | This is a feature the harness vendors will (and Anthropic already did) ship natively after each incident, leaving a third-party cross-agent proxy as a fragile shim enterprises would rather cover with existing DLP. |
| 4.2 | SafeSearch policy layer for managed networks | NextDNS and CleanBrowsing already enforce SafeSearch and block non-enforceable engines at $20-40/yr; keeping a registry of new engines current is a config update for them, not a company. |
| 4.1 | LLM-judged audit of past agent sessions | LLM-as-judge over agent traces is already a commoditized checkbox in LangSmith, Langfuse and Braintrust, so this reduces to a prompt template and a corpus claim with thin evidence (one commenter 'implying demand'). |
| 3.9 | Ad-free school search with enforced SafeSearch | Districts can default to Bing/DuckDuckGo with SafeSearch enforced by the Securly/Lightspeed filters they already own for free, Kagi could add DNS enforcement as a single feature, and building a competitive search index is a capital sink. |
| 3.6 | GPU parameter UI kit for audio plugins | The entire evidence base is one person's abandoned Python/Tk project plus a wish, and iPlug2 already gives indie audio devs a free, permissive plugin UI framework they can skin with GPU rendering. |
| 3.4 | Permissive GPU desktop UI toolkit for Python | DearPyGui is already a free, MIT-licensed, GPU-rendered Python UI toolkit, and building a full widget toolkit with bindings is a multi-year effort an indie cannot recoup at $100-300/yr from hobbyist tool authors. |