What I’m Building: September 2026

Image is merely illustrative.

A few hundred builders have followed me recently, so I thought I should compile and share what I’ve been building.

My startup, Socra, is a human-AI augmentation platform. We augment builders and AI agents with knowledge, tools, and shared context.

I think that by 2050, the median company will have only two humans. People will run their own businesses with teams of AI agents, doing work that takes thousands of engineers today. This is what I mean by “everyone will be a developer.” We’re building Socra for that future, starting with ourselves because we’re that team too (Our team has only two humans: Mike and me). Every tool we build should make us and our agents more capable. Every lesson we learn should improve future work. That’s how we intend to build hundreds of products with only two humans, and make the tools that make our work compound available to other builders.

We’re building Socra in a way that scales and compounds. Every new tool increases what we can make. Each product adds parts we can reuse, reducing the cost and effort of building the next one. When we improve a shared service, the products using it benefit. When we improve a UI component, our products get that improvement when they update the shared library. Our agents carry those improvements across products, so the work we do in one place can improve many others.

A new product already has somewhere to run, a way to sign people in, billing, and components for its interface. We can spend our time and tokens on what makes that product useful and unique.

Cortex lets us reuse the knowledge behind what we build too. Knowledge is the layer where we operate now that our agents write the code. We define how our systems should work and why. Our agents load those decisions before they start working.

Each subject has one source of truth, a Module. Modules declare dependencies on the knowledge needed to understand them, forming a Semantic Directed Acyclic Graph, or SemDAG. A flash loads the relevant Module and its dependencies, with the foundations first and each Module included once.

When we learn something that changes a decision, we update its Module. The next agent that flashes it gets the new version, even if it’s working on another product in another tool.

We no longer duplicate that knowledge in agent memories, skills, or AGENTS.md files. Keeping separate copies meant more work to maintain them, conflicting versions, and repeated/uneeded context consuming tokens. With Cortex, we maintain the knowledge once and our agents load it wherever it’s needed. Mike explains the idea in Agent Context Needs an Import Statement.

Here’s what we’ve built, organized by product line.

• Cortex: Stop Repeating Yourself to Your Agents. Available.

• Cortex Issues: Work connected to the knowledge that governs it, with dependencies, assignments, comments, and attachments. Available.

• Cortex Inbox and schedules: See which work is ready and schedule recurring tasks. Available.

• Cortex MCP and agent plugins: Share Context Across Your AI Agents. Connect through plugins, MCP, or the CLI. Available.

• Journeys: Augment Your Thinking with AI Agents. Available.

• Channels: Keep the Conversation Moving. Available on Mac.

• Admin: Manage Your Workspace. Available.

• Fab: My software factory for building and improving software with agents. Internal right now; will become available soon.

• Cortex and Channels mobile apps: Knowledge, tasks, and conversations on your phone. In development.

Access depends on your Account and Project.

• Domains: Verify domain ownership and claim hostnames.

• Build: Turn source code into an immutable OCI artifact.

• Run: Deploy workloads and manage their revisions.

• Container Registry: Store and distribute Project-owned OCI images.

• Sandbox: Run commands in durable cloud environments.

• Databases: Provision isolated, managed PostgreSQL databases.

• Secrets: Manage write-only runtime secrets and their versions.

• Pub/Sub: Publish messages and consume them through subscriptions.

• OAuth: Register clients and authorize application access.

• IAM: Control access to Project resources with policies.

• Analytics: Collect events and query reports in datasets.

• Service Usage: Enable services and inspect usage for a Project.

• Projects: A home for your application’s resources, access, and service usage.

• Cloud Console: Manage Cloud projects and services through the browser.

• Cloud agents: Agents with their own identity, persistent cloud environment, mailbox, and access to the repositories they work on. Coming soon.

• Email for builders and agents: Manage contacts, subscription preferences, templates, campaigns, and sending domains. Coming soon.

• Source: A cloud-native source system with workspaces, snapshots, and verified changes. In development.

• Work: Give agents ongoing responsibilities and track the work they perform. In development.

These systems support our products. They are all internal right now and some may become available in the future:

• Identity: One personal identity across Socra, independent of your email address.

• Accounts and Directory: Organization membership, groups, and identities for people and agents.

• Billing: Subscriptions, invoices, prepaid credits, automatic refills, and usage charges. Stripe handles payment collection.

• Licensing: Product access and seat assignments, managed separately from billing.

• API framework: Define a service contract once and share it between the server and its clients.

• MCP adapters: Expose service operations as tools agents can call.

• MCP toolkit: Define MCP tools and servers in TypeScript.

• ORM: Define database models, query data, and manage migrations.

• Pkg: Create packages, request changes from agents, and publish releases through our npm-compatible registry.

• Git hosting: Private repositories with permissions for people and groups.

• Git Store: Git repositories that other services can create and manage.

• Volumes: Persistent cloud storage with snapshots and forks.

• Mail: Mailboxes and native SMTP delivery, with incoming mail, outgoing queues, and quarantine.

• socra.design: Zero Friction. The design system behind every Socra product. Public catalog; internal libraries.

• Socra CLI: One CLI for Everything Socra. Available.

• Shared editor: Rich-text editing reused across our applications. Internal.

• Shared web foundation: Common authentication, layouts, documentation, and marketing components. Internal.

• Icons and graphics: One source for the vectors used across our products. Internal.

• Browser automation: Browser environments and MCP tools for agents to inspect and operate websites. Internal.

• Agent host tools: Controls for running agent sessions on our machines. Internal.

• Repository and release tooling: Create repositories, run checks, version packages, and publish releases. Internal.

• Video tooling: Produce demonstrations using the same interface components as our products. Internal.

• SemDAG: The Semantic Directed Acyclic Graph that connects Modules by dependency. Public reference.

• Documentation: Product guides and references for people and agents. Public.

• Support: Need help? Start here. Public.

• Support publishing: Maintain and publish the answers in our help center. Internal.

• eduarda.ai: How I use Cortex to augment my AI agents. Public.

• Growth: Research how people and agents discover Cortex, track citations, and run experiments. Internal.

I’m probably leaving some things out. We’re building a ton! What excites me most is how each improvement carries into the next thing we build. We keep getting faster, and the work keeps getting better.