Gameplan
Gameplan replaced its generic project-management SaaS with a custom client portal, an embedded AI assistant named Ari, and an architecture designed for an agentic future where the UI itself becomes optional.
Client
Darren · Gameplan Technologies
Marketing services agency
Engagement
Fractional product team
Surface
Portal · AI layer · Admin
Status
In production
2026 — ongoing
The problem
Gameplan is a marketing services agency. The business sells brand work, design, and ongoing execution to clients who don't have a marketing team of their own. To manage all of it, Darren had stitched together Assembly — a generic project-management SaaS — to handle client requests, deliverables, and team workflows.
It worked. Until it didn't. Assembly cost real money every month, it was built for everyone (which means it was built for no one), and every product decision Darren wanted to make for his agency was bottlenecked by whatever Assembly's roadmap felt like shipping that quarter. Worse, the v1 of the actual Gameplan app was running on Bubble — a no-code platform with limits Darren had hit years earlier.
The mandate was clear: replace Assembly. Replace Bubble. Build a portal Gameplan actually owns — fast enough that the business doesn't skip a beat, and architected so it can absorb whatever AI does to agency software over the next three years.
The thesis
Most agency-portal rebuilds stop at "we have a Kanban board now." Gameplan's didn't. The architecture was designed in three layers from day one:
v1. A working portal — services catalog, Kanban, file library, client and admin views, Stripe checkout, email notifications. The "we own this now" baseline.
v2. Ari — an embedded AI layer that intakes new client work conversationally, drafts task briefs, scopes services, books meetings, and orchestrates external tools like Canva for design output. The portal stops being a place to fill out forms and starts being a place where work gets assigned through conversation.
v3. A faceless MCP-native brand-context layer. The end state where a client never opens the portal at all — they ask Claude (or any AI agent) for design help, the agent calls Gameplan for brand context, and the work happens elsewhere. Gameplan becomes the intelligence hub, not the UI.
The discipline is shipping v2 while writing v3-compatible code. Every Ari capability is built as a skill that can later be exposed via MCP. The portal is the wedge. The brand-context platform is the moat.
v2 is Gameplan telling Canva what to build. v3 is the user asking Claude, and Claude asking Gameplan for the brand context. The portal becomes optional.
— Product strategy note, Apr 2026
The build
The team executed three parallel workstreams without breaking client service.
The migration. Bubble v1 → Next.js v2 on a fresh stack. Supabase replaced the legacy Flusk security layer. Stripe replaced manual invoicing. Resend handled transactional email and team invites. Vercel preview branches gave Darren a way to review every PR before it merged. Assembly's role got absorbed into a custom Kanban + file library inside the portal.
The AI layer. Ari started as a sidebar chat scoped to individual service pages and grew into a full-screen workspace. The "loop in humans" feature turned every Ari conversation into a 3-way thread where a Gameplan strategist could jump in mid-flow. Ari got skills for brief intake, service scoping, task writing, and Google Calendar booking. Every skill is structured to graduate to an MCP tool when v3 arrives.
The Apps Framework. Discrete capabilities — Website Assessment, Competitor Analysis, Logo Creator — built as installable "apps" inside the portal, each with its own scope, billing hook, and Ari skills underneath. The architecture: apps are the contract, skills are the implementation.
Next.js 15 · App Router
Portal · client + admin
TypeScript · Tailwind
Type-safe · design system
shadcn/ui
Component base
Supabase
Postgres · Auth · Storage
Vercel AI SDK · Claude Sonnet
Ari runtime
Stripe
Checkout · subscriptions · custom cart
Resend
Transactional · invites
Google Calendar API
OAuth · Ari scheduling
Execution
This was never a "build it and disappear for three months" project. The cadence was deliberate: a working call with Darren most weeks, Loom walkthroughs of every meaningful PR, Slack for daily back-and-forth, Notion for the task board that both sides could see and reshape in real time.
The reconciliation discipline mattered more than the velocity. Before every check-in, the Granola meeting record from the last call got cross-checked against shipped PRs, the Notion task board, and Slack threads — so we never showed up to a call uncertain about what was done. When Darren asked for changes mid-stream (and he did), they got captured as discrete tickets, scoped, and prioritized openly rather than absorbed silently.
Multiple times during the build, Darren pivoted the product direction — most significantly when the Apps Framework emerged in April. The architecture absorbed it without a rewrite because the foundation had been built for change.
3
Platforms migrated off
70+
PRs shipped to date
35+
Notion tasks closed
Weekly
Client cadence
Timeline
Feb 2026
Foundation
GitHub access, V0 code audit, Supabase rebuild from scratch. Auth, schema, and the existing UI wired to a real backend. Kanban, file library, services catalog brought online.
Mar 2026
Assembly out, Stripe in
Custom domain cutover. Stripe sandbox → live, custom cart, embedded checkout. Resend integration for invites and onboarding emails. First client work runs through the new portal end-to-end.
Apr 2026
Ari, full-screen
Sidebar Ari expands into a full workspace. 'Loop in humans' 3-way chat. Google Calendar P1 — OAuth, availability lookup, Ari-driven booking. Sparkle-icon UX, service-page polish, auth + invite hardening.
May 2026
The Apps Framework
The v2/v3 pivot lands. Capabilities ship as installable apps — Website Assessment, Competitor Analysis, Logo Creator — each one a contract over Ari skills underneath. Logo Creator POC scoped for an investor demo; Canva integration scaffolded for orchestration.
What it proved
Software that fits the agency
Assembly served everyone, which meant it served no one well. The new portal does exactly what Gameplan does — services, briefs, deliverables, Kanban, client view, admin view — and nothing else. Every feature has a job.
An AI layer the business owns
Ari isn't a third-party widget. It's a custom assistant trained on Gameplan's services, brand context, and intake patterns, with skills that compose into apps. No vendor can take it away or change the price.
Architecture built for what's next
Building v2 while writing v3-compatible code means the MCP-native future isn't a rewrite — it's a migration path. When agency UIs collapse into agent calls, Gameplan is ready.
A working relationship, not a transaction
Weekly cadence, transparent backlog, scope discussions in the open. The build still ships when priorities pivot — because the working agreement is designed for change, not for fixed-scope delivery.
Stuck renting software that doesn't fit your business?
