Inspiration
We started by asking what software could do after it was built.
During ideation, we discussed a friend commercializing proprietary mathematical IP. A speculative idea about software protecting itself from reverse engineering led us toward a different question: could software recognize where else it would be useful and help its creator distribute it?
We followed that question from teammates to departments, then across companies. Coding agents are making bespoke software easier to produce, but useful tools still get buried in repositories, private conversations, and individual workflows. Someone else ends up rebuilding the same thing.
Social Capital inspired the portfolio-level opportunity. Investment funds connect companies through capital, expertise, and relationships. What if the software created inside those companies could benefit from that network too?
Common Ground grew from that idea: a software factory whose output can reach the next team that needs it.
What it does
Common Ground builds, checks, and shares useful software through a VS Code-compatible interface.
An engineer describes a supported task. A coding agent produces an implementation, runs independent checks, and packages the checked version. Common Ground recommends potential recipients using shared workflow briefs, recorded reuse, and permission-aware retrieval. The author reviews the evidence and approves the share.
The recipient opens the project in their editor and runs the shared code with their own inputs.
Our demonstration explores an illustrative connection between Saildrone-inspired ocean survey operations and Palmetto-inspired solar servicing. Their businesses are different, but both can benefit from tools for checking measurements or planning weather-dependent fieldwork.
These are demonstration scenarios, not customer deployments or claims about either company’s internal systems.
How we built it
We built a VS Code-compatible extension, a hosted account and sharing service, and a local generation and execution pipeline.
- Claude Code generates the supported telemetry inspector using a blueprint and independent behavioral checks.
- 8090 records requirements and provides a reference in the artifact’s build lineage.
- HydraDB retrieves shared workflow context within permission-eligible company scopes. Workflow keywords and recorded reuse provide a fallback.
- Upstash Redis stores organizational state, memberships, tool packages, and sharing grants.
- QuickJS executes supported tools locally in a restricted environment.
- ElevenLabs Scribe supports an optional recording-to-brief input path.
- Vercel hosts the website and sharing API.
Our initial ontology connects funds, companies, people, permissions, workflow needs, input contracts, and reuse relationships. Recommendations identify potential value; explicit grants control access. Source hashes identify the exact code being shared and executed.
Challenges we ran into
Identity and permissions were harder than they first appeared. One person can administer a fund while also working inside a portfolio company. We needed to distinguish those roles, keep company selection consistent across laptops, and prevent stale state from showing the wrong workspace or shares.
We also had to separate relevance from authorization. A tool may be useful to someone who has opted out. A recommendation must never grant access by itself.
Generated code introduced another challenge: plausible output is not enough. Our inspector checks cover missing observations, duplicates, malformed values, valid zeros, and incompatible assumptions.
Finally, we worked to keep the developer experience understandable. Receiving, saving, and executing a project are distinct actions, but they need to feel like a natural continuation of work inside the editor.
Accomplishments that we're proud of
We completed a real two-person handoff: a tool built on one laptop was shared, opened, and executed on another against a different company’s sample data.
The shared inspector runs against public ocean and solar datasets while preserving the core source code. Its checks distinguish actual missing observations from legitimate measurement values.
We also built four runnable reference projects: weather-window planning, revenue reconciliation, battery-test review, and inference-capacity planning. Each has editable inputs and project-specific acceptance checks.
Our selective-routing rehearsal demonstrates that the system does more than broadcast: changing a company’s workflow brief changes its recommendation eligibility, while opting out overrides relevance.
Most importantly, the prototype connects creation, verification, discovery, approval, and execution into a working flow.
What we learned
Producing software is only part of realizing its value. Discoverability, compatibility, trust, and delivery determine whether another person can actually use it.
We learned that a useful ontology starts with explicit relationships and contracts. Knowing which company someone belongs to, what its team needs, and what inputs a tool accepts can already enable meaningful reuse.
We also learned to keep our claims precise. Our live generation pipeline currently specializes in telemetry inspection. The four broader projects are prepared, tested examples of the sharing and execution system. Matching currently relies mainly on company-level needs rather than a complete understanding of each engineer’s work.
What's next for Common Ground
We want to deepen recommendations from company-level briefs to individual workflows, expand the kinds of tools the factory can generate, and strengthen independent verification of shared packages.
We also want to support dependencies, version updates, and richer adaptation between different teams’ environments while keeping the interface focused.
Our ambition is that every useful tool a company builds has a path to its next authorized user, and that each software factory makes the rest of the portfolio more capable.
Built With
- 8090
- claude
- css
- elevenlabs
- extension
- html
- hydradb
- javascript
- node.js
- quickjs
- redis
- typescript
- upstash
- vercel
- vscode