What one person actually built

Jesse James Richard
|
Aug 3, 2026
|
15 min read
#Signature
#Architecture
#AI & Agents

A normal company assembles a team and thirty-odd SaaS subscriptions to run a platform like this. This one is one person, one integrated system, built from the ground up: 474 operations across about thirty scopes, roughly 180 packages, 14 cloud services, four platforms. What follows is every piece of it, and how it fits together.

0

API operations

~0

API scopes

~0

packages

0

cloud services

0

platforms

The tenancy spine

Multitenancy. Isolates every organization's data, configuration, and sites on one shared platform, so no tenant ever sees another. It is the precondition for everything above it, and it is what makes the agency market possible: one operator running thousands of sites. Multitenant architecture, tenant isolation, an organization-project-app hierarchy at scale. Replaces: the multitenant framework a team otherwise stitches together. Built on: Postgres, Cloud Storage.

Authentication. One identity a person carries across web, desktop, mobile, and the API, with email, social sign-in, and a desktop flow. Every surface uses the same account, so a person signs in once and is known everywhere. Identity, sessions and tokens, federated sign-in, cross-platform auth. Replaces: Auth0, Clerk. Built on: Identity Platform, Postgres.

Role-based access control. Governs what each user and machine may do, down to a single permission, because a multitenant multi-role platform cannot have coarse access. It is the trust and compliance layer enterprises want before they will buy at all. Authorization, fine-grained access control, least privilege, roles and scopes. Replaces: a permissions vendor (WorkOS, Oso). Built on: Postgres.

Invitations. Brings a new person in by token: they arrive with context and are granted a role in one step, not a support ticket. It is the seat-expansion mechanic every team product needs. Token flows, seat-based onboarding. Replaces: the invite flow inside a team-management tool. Built on: Postgres.

Team membership. Membership at the organization level and again inside each app, because belonging is not all-or-nothing: someone can be in the org but only some of its apps. The seat and role granularity enterprises expect. Role hierarchies, scoped membership. Replaces: the member-management piece of a team platform. Built on: Postgres.

Service accounts. Machine identity with the same granular grants as humans, through the same stations, no special doors, so automations and outside systems act with scoped, auditable access. The enterprise and extensibility layer. Machine identity, non-human principals, scoped grants, auditability. Replaces: a machine-identity vendor. Built on: Postgres, Identity Platform.

API keys. Personal and organization keys for machine callers, since the API is a product surface and callers need durable, revocable credentials. A developer-extensibility surface, and a monetizable one. Credential issuance, revocation, programmatic access. Replaces: an API-key management layer. Built on: Postgres.

The infrastructure it runs on

The database. Postgres, after six years on a document store: the single source of truth for every organization, project, app, and piece of content. The move was foundational, because the platform needed relational, multitenant queries the old store could not give it. Relational modeling, migrations, connection pooling, multitenant schema design. Replaces: a managed database service. Built on: Postgres, Cloud SQL.

Object storage. Holds files, media, and public site assets in private and public buckets, because the platform generates and ingests large binaries that do not belong in the database. Object storage, public and private access, per-tenant paths. Replaces: a stored-assets product. Built on: Cloud Storage.

Durable jobs. Runs long, asynchronous work that survives restarts and spreads across workers, because the AI jobs run for minutes and cannot die on a deploy. Work queues, idempotency, at-least-once delivery, horizontal scale. Replaces: a managed queue service. Built on: Redis.

Serving and caching. Renders and serves hundreds of tenant sites fast from one stack, with server-side rendering and a caching layer, so every tenant site is quick without a per-tenant deployment. Multitenant serving, server-side rendering, caching and invalidation. Replaces: what you would rent from Vercel or Netlify plus a CDN like Cloudflare. Built our own on the raw cloud instead. Built on: Cloud Run, Cloud Load Balancing.

Custom domains. A customer connects their own domain and the platform handles the DNS, the load-balancer address, the certificate states, and teaching the router to resolve it, so every tenant serves on its own brand. DNS, TLS certificate lifecycle, domain-to-tenant resolution. Replaces: the custom-domain feature of a site host. Built on: Certificate Manager, Cloud Load Balancing.

Routing. Resolves every request to the right tenant, app, and view, on a routing layer built in-house because no off-the-shelf router fit the multitenant, multi-app, multi-platform shape. Client and server routing, code-splitting, deep links. Replaces: the framework routing you take from a hosting platform. Built on: Cloud Run.

Secrets. Holds credentials and configuration secrets across every service in one secure place. Secret storage, rotation, least exposure. Replaces: a secrets-management vendor. Built on: Secret Manager.

The API and code generation

The API. An API-first platform of 474 operations across about thirty scopes, where the API surface is the product's true feature map: everything a person or machine can do is an operation on it. API-first design, resource modeling, naming, versioning. Replaces: the hand-built API layer a backend team maintains. Built on: Postgres, Cloud Run.

Code generation. One schema is the source of truth, and the clients, the types, the permission definitions, and the SDK are all generated from it, so no two surfaces can drift apart. This define-once discipline is the backbone the platform's consistency rides on. Schema-driven code generation, single source of truth. Replaces: an SDK-generation vendor. Built on: the platform's own schema pipeline.

The SDK. A generated client library so outside developers call the platform in their own code, kept in lockstep with the API by the same generation step. Client SDKs, developer ergonomics. Replaces: an SDK-as-a-service. Built on: the code-generation pipeline.

MCP. The platform describes its own routes and exposes them as tools an AI model can call, so agents operate it the way a developer would. Self-describing interfaces, machine tool protocols. Replaces: a bespoke integration layer. Built on: the API schema.

The developer portal. A full developer surface, twenty-seven operations of it: keys, documentation, the SDK, the MCP tools, everything an outside builder needs to build on the platform. Developer experience, API documentation. Replaces: a developer-portal product. Built on: the API and code generation.

The builder

The website builder. The largest scope on the platform, fifty-six operations: it composes a business's pages, blocks, layouts, headers, footers, and menus into a live site. Structured content, composition, publishing. Replaces: Webflow, Framer. Built on: Postgres.

The design system. One block model that renders identically wherever it appears, one place where the design rules live, and a theming layer, so a business's site stays coherent no matter what generates it. Component models, design tokens, theming. Replaces: a design-system team. Built on: Postgres.

The editor. A full rich-text and block editor, the builder's hands: the surface a person or the platform edits content through. Rich-text editing, structured documents, block composition. Replaces: an embeddable editor product. Built on: a rich-text editor engine.

Content stages. Content moves through stages, from what a page should say to how it should look, each a separate step the platform can reason about on its own. Pipeline design, separation of concerns. Replaces: the editorial workflow a team runs by hand. Built on: the agentic pipelines.

Visual QA. The builder screenshots its own rendered output and inspects it, so it catches what a rendered page gets wrong before a person ever sees it. Automated visual inspection, headless rendering. Replaces: a visual-regression tool. Built on: a headless browser.

Localization. The platform and every tenant site run in three languages, with translation carried through the same access rules as everything else. Internationalization, translation management. Replaces: a localization platform. Built on: an i18n framework.

SEO. Every tenant site gets its sitemaps, its search-index pings, and its search-console registration automatically, so being found is a platform feature, not a chore. Technical SEO, sitemap generation. Replaces: an SEO tool and the person who runs it. Built on: Search Console, IndexNow.

Branding. A business's brand is defined once and cannot drift: the same voice, palette, and rules govern every page, email, and post the platform makes. Brand systems, single source of truth. Replaces: a brand-management tool. Built on: Postgres.

The AI and autonomy layer

The Mind. The autonomous orchestrator, the third-largest scope on the platform: it decides what content a business should have, builds it, and drives it out. This is the differentiator with no vendor category to replace, because nothing else does it. It is what moves the product from a website tool to an autonomous marketing platform, and what the pricing is justified on. Autonomous orchestration, planning, agentic workflows. Replaces: nothing on the market. Built on: the durable job runtime, the model, Postgres.

Durable agentic pipelines. The substrate the Mind runs on: a contract holds each job's shape, work survives restarts, and a bad step is fixed in place instead of re-rolling the whole thing. Without it there is no reliable autonomy above it. State machines, checkpointing, fault tolerance, model-call resilience and reuse, horizontal scale. Replaces: an agent framework. Built on: Redis, the model.

Fact extraction and grounding. One pass extracts the facts from a business's material, and a second generates content grounded against those facts, so nothing is invented. It is the trust layer that makes autonomy sellable, the difference between generic AI text and a business's own content. Retrieval-augmented generation, fact extraction, hallucination control. Replaces: nothing generic offers it. Built on: the model, Postgres.

Embeddings and retrieval. Turns files and content into vectors the platform can retrieve and cite, so it finds and reuses a business's own material by meaning rather than keyword. Vector search, semantic retrieval. Replaces: a vector database and search vendor (Pinecone, Algolia). Built on: Postgres, the model.

The learning loop. The accept button captures training data, so the platform improves from its own approvals with no separate labeling effort. Every customer's approvals make the product better. Human feedback, training-data capture, continual improvement. Replaces: a data-labeling pipeline. Built on: Postgres, BigQuery, the model.

Multichannel autonomy. Web, mail, and social from one mind, grounded in one understanding of the business, so one voice reaches people across every channel. It is the headline product story, the marketing department a business never hires. Replaces: the whole web, email, and social stack a marketer coordinates. Built on: the Mind, the apps below.

The apps built on the core

The knowledge base. A business's knowledge, written and maintained by the platform and searchable by meaning, twenty-one operations of it. Knowledge management, retrieval. Replaces: a help-center product. Built on: Postgres, the model.

AI chat. A chatbot that answers from the business's own knowledge, grounded in what it has ingested rather than what it guesses. Retrieval-augmented chat. Replaces: a support-chat vendor. Built on: the model, Postgres.

Files. Upload, storage, and lifecycle for a business's documents and media, twenty-three operations, and the raw material the mind reads to ground everything else. File handling, ingestion. Replaces: a file-management service. Built on: Cloud Storage, Postgres.

Media processing. New files are compressed and resized, video and audio are transcribed, images and documents have their text pulled out, so a fact stated once in any file is usable everywhere. Media pipelines, transcription, optical character recognition, document extraction. Replaces: a media-processing and transcription vendor. Built on: Speech-to-Text, Vision, Cloud Storage.

CRM. A contact record and activity log written for a model to read, twenty operations, so the platform knows who each person is and what they have done. Customer data, activity tracking. Replaces: HubSpot, Salesforce. Built on: Postgres.

Forms. Forms that cannot drift, whose submissions become contacts in the CRM, so capturing a person and later writing to them draw on the same source of truth. Form building, structured capture. Replaces: Typeform. Built on: Postgres.

Outbound email. Mind-driven email with no campaigns or segments: the platform decides who to write to and when, one recipient at a time, twenty-eight operations. Transactional and lifecycle email, per-recipient planning. Replaces: Customer.io, Mailchimp. Built on: Postgres.

Notifications. A delivery and preference layer every channel dispatches through, honoring what each person agreed to receive. Notification delivery, preference gating. Replaces: Knock, Courier. Built on: Postgres.

Socials. The registry of a business's social accounts, the single source of truth the builder reads its social links from and the channel the mind will post to. Account registry, channel identity. Replaces: the account side of a social tool. Built on: Postgres.

Data and analytics

Analytics. Usage flows into the warehouse and back into the platform, so the same data that shows a tenant their numbers also informs what the mind decides to build next. Product analytics, event pipelines. Replaces: Segment, Amplitude. Built on: BigQuery.

The warehouse. BigQuery holds the data that serves two masters at once: the dashboards a tenant sees and the decisions the mind makes. Data warehousing, analytical queries. Replaces: a data-warehouse and pipeline stack. Built on: BigQuery.

Observability. Distributed tracing where every request carries its own cost, so a trace and its spend are one event rather than two systems to reconcile. Tracing, cost attribution. Replaces: Datadog. Built on: BigQuery.

Billing. Usage metered in units with flat pricing, built on a payments rail rather than a subscription vendor, so the platform owns its own metering and pricing. Metering, usage-based billing. Replaces: Chargebee, Recurly. Built on: a payments rail, Postgres.

Operations and reliability

Continuous delivery. An expansive pipeline: branch tests, deploy on merge, an auto-release lane, a flow that enforces how changes reach production, and a hotfix lane, all running the test suite in the cloud. Continuous integration and delivery, release engineering. Replaces: a CI vendor and a DevOps hire. Built on: Cloud Run.

Tests. Tests treated as part of the design, with an infrastructure and a cache fast enough to run them on every change, learned the hard way through a suite that kept dying. Test design, coverage, fast feedback. Replaces: a testing platform. Built on: the toolchain.

Local development tooling. The dev environment: hot reload across services and a team of agents working in visible panes, so one person operates like several. Developer tooling, orchestration. Replaces: a platform-tooling team. Built on: the toolchain.

Auto-healing. A service that catches production errors and comes back with a diagnosis and a fix attached, so failures arrive already half-solved. Error tracking, root-cause analysis, self-healing. Replaces: Sentry and the on-call who triages it. Built on: Postgres, the model.

Bug reports and feature requests. In-product channels for users to report a bug or ask for a feature, routed into the auto-healing service that reads and diagnoses them. Feedback capture, triage. Replaces: Canny, Linear. Built on: Postgres.

Legal documents. Terms and policy documents managed and standardized in the platform, versioned and served per tenant. Document management, compliance. Replaces: a legal-ops tool. Built on: Postgres.

Onboarding. The path that takes a new business from nothing to a running site, wired into invites, apps, and the builder. User onboarding, activation. Replaces: an onboarding tool. Built on: Postgres.

Trash and soft-delete. Deleted things go somewhere recoverable rather than vanishing, so a mistake is reversible across every kind of content. Soft deletion, recovery. Replaces: a feature teams rebuild per app. Built on: Postgres.

Four platforms. The same product runs on the web, as a server-rendered site, as a desktop app, and as a mobile app, from one codebase. Cross-platform delivery. Replaces: separate web, desktop, and mobile builds. Built on: the shared client.

The apps model. One app per deliverable, a framework every capability above plugs into, so the platform grows by adding apps rather than rewiring itself. Modular architecture, extensibility. Replaces: the framework a platform team maintains. Built on: the core.

What it replaces

Read one more way, as a purchase order. Each row is a product a company would otherwise license, and one of the things built here in its place.

Capability
Bought here as one system, instead of

Authentication

Auth0, Clerk

Access control

WorkOS, Oso

Website builder

Webflow, Framer

Serving and CDN

Vercel, Netlify, Cloudflare

Vector search

Pinecone, Algolia

CRM

HubSpot, Salesforce

Forms

Typeform

Outbound email

Customer.io, Mailchimp

Notifications

Knock, Courier

Analytics

Segment, Amplitude

Observability

Datadog

Billing

Chargebee, Recurly

Error tracking

Sentry

Feedback and issues

Canny, Linear

The technologies, seen the other way

The same platform, read as a cross-reference: each piece of infrastructure and where it earned its place.

Cloud Run runs every service. Cloud SQL holds Postgres: all platform data, the funnel state, the knowledge and vector store. Cloud Storage keeps files, media, and public site assets. BigQuery is the warehouse feeding both the analytics a tenant sees and the decisions the mind makes. Identity Platform backs authentication. Certificate Manager and Cloud Load Balancing carry custom domains. Secret Manager holds secrets across every service. Speech-to-Text transcribes audio and video. Vision reads text from images and documents. Search Console and IndexNow drive per-tenant SEO. The generative model, described by what it does rather than named, powers the Mind, grounding, embeddings, chat, and the builder.

Above the infrastructure sit the capabilities built in-house: a rich-text editor, a durable job queue, an image and media pipeline, a schema-to-code generation chain that produces the API, SDK, MCP tools, and permissions from one definition, an internationalization layer, and the client that renders across four platforms. None of it is a rented service. All of it is one codebase, in one language after a polyglot era.

What it adds up to

This is not thirty-five features in a pile. It is one platform where every piece knows about the others: the tenancy spine authorizes everything above it, one schema generates every surface, the Mind orchestrates the apps, and the data loops back to make the model better. A company assembles a team and thirty vendors and pays people to keep the seams from showing. This is one person, one integrated system, with no seams because there was only ever one builder. That is the argument: build all of this, and there is not much left you cannot build.

That is the argument: build all of this, and there is not much left you cannot build.

The case for building it, and the market it opens, is where this series began.

Have questions?

If you're interested in my work or Giant Context, contact me!

Contact Jesse
Legal
Privacy Policy
Terms of Service
Cookie Policy
Hosted on
What one person actually built | Jesse James Richard