Senior Full-Stack Engineer
Next Best Action

What you will build

A multi-tenant Next.js application where IG's clients, mid-market CMOs and founders, log in to see live engagement status, ask natural-language questions of their own marketing data with sourced answers, and approve agency work with one tap. The agent runs on the Claude Agent SDK with tools wired into HubSpot, Salesforce, GA4, Google Ads, Meta Ads, RudderStack, and tenant warehouses (Postgres, Snowflake, BigQuery, read-only).

The MVP ships in eight weeks to a single live pilot. The next eight weeks add four more pilots and harden the product. The twelve-month arc takes Next Best Action to eight to twelve paying clients on $3K to $20K per month workspace pricing. You are the second most important person in the company on this product.

This is not a moonshot. The full strategic, architectural, and pricing spec is written. You will read it on day one and start contributing to it on day two.

What you will own

• The agent harness. Claude Agent SDK in TypeScript, sandboxed runtime, per-session caps on turns, tokens, and cost. Tool allowlist composed from tenant, role, and task.
• The integration layer. Thin Tool abstraction with MCP and HTTP backings. HubSpot MCP, Salesforce MCP (April 2026 GA), RudderStack MCP, plus custom in-process tools for GA4, Google Ads, Meta Ads, and read-only warehouse queries.
• The multi-tenant pattern. Three layers of defense: auth boundary, RLS on every tenant-scoped Postgres table, namespace isolation on the playbook RAG. Tenant context is bound at the request layer and never accepts user input.
• The Next.js app. App Router, React Server Components, streaming SSR. streamText from Vercel AI SDK for the Ask surface. Tailwind and shadcn/ui for the design system. Clerk for auth, Neon for Postgres, Upstash for Redis.
• Cost and observability. Per-tenant spend tracking, anomaly detection, token-budget caps, audit log on every tool call. OpenTelemetry traces, Helicone for LLM-specific spans.
• The first eight to twelve weeks of the product. You ship features, you fix bugs, you respond to Customer Zero feedback, you write tests, you hold the quality bar. There is no QA team yet. There is no DevOps team yet. You will lean on Vercel and Neon for the operational primitives and on Chris and the IG team for the product calls.

What we are looking for

You have

• Shipped a multi-tenant SaaS product before. Postgres with RLS, OAuth-per-tenant secret handling, tenant-scoped audit logs. You know what goes wrong and how to prevent it.
• Worked with LLM agents at production scale. Claude Agent SDK, OpenAI Assistants, LangChain LCEL, or equivalent. You have shipped tool-using agents that called real APIs against real data and you know the failure modes. Prompt caching, structured outputs, tool errors handled in-band, latency versus cost trade-offs.
• Strong Next.js 15 / App Router experience. Server Actions, RSC, streaming, Vercel AI SDK. You can build a product UI without a designer holding your hand and without splitting the role across two engineers at MVP.
• Comfort with the trade-off between speed and quality. You know when to ship V0.0 in three days and when to slow down for the hardening pass. You will not gold-plate v1 or skip migrations.
• Communication that respects context. You can read a 200-page spec and ask the four questions that matter, not the forty that are answered if you keep reading. You can write a six-line PR description that helps a non-engineer reviewer understand what changed.

You probably have

• TypeScript fluency including type-driven schema design with Zod, Valibot, or drizzle-zod.
• Experience with vector stores (pgvector, Pinecone, Turbopuffer) and RAG patterns.
• Familiarity with MCP (Model Context Protocol), first-party servers and custom in-process tools.
• Knowledge of B2B SaaS auth patterns: SSO, SAML, organizations, role-based access. Clerk and WorkOS specifically.
• Comfort working alongside a non-technical product owner. You will not be reviewed on velocity alone. You will be reviewed on whether Maggie at the pilot can do her job better.

You do not need

• A specific framework allegiance. We chose Next.js for the ecosystem. If you have a strong case for Remix or Astro, we listen, but the default is Next.
• Mobile development experience yet. Capacitor wrap is post-MVP.
• DevOps or SRE background. We are on Vercel and Neon for the foreseeable future.
• A computer science degree.

What you should expect from the process

1. Application. A brief written response to: "Walk me through a multi-tenant SaaS bug you debugged in production. What did you change, and what did you almost change but did not?" 200 to 400 words.
2. Conversation with Chris (45 min). The bet, the spec, your reaction. We will share the spec and the mockup ahead of time. Come with questions and pushback.
3. Architecture review (90 min). We walk through the technical architecture spec together. You tell us what you would change. This is the load-bearing screen. We are looking for clear thinking, not the right answer.
4. Reference check. We will ask to talk to one engineer who reported to you (or a peer who collaborated closely) and one product owner who worked with you.
5. Offer. Within five days of the architecture review.

We do not do whiteboard interviews. We do not do trick algorithm questions. We are hiring you for what you will build, not for what you can recall.

What success looks like

• Day 7. Repo bootstrapped, CI/CD live on Vercel, Clerk Organizations wired with one test workspace. You have read the spec and given Chris three corrections that improved it.
• Day 14. V0.0 demo. You can sign in as Maggie, ask a HubSpot question against a sandbox, and get a streamed answer. Cost is tracked in Postgres.
• Day 30. Three screens (Today, Status, Ask, Approve, four counting Today) work end-to-end on hard-coded data. RLS in place. The IG admin console exists in skeleton form.
• Day 60. Customer Zero ships. Maggie has a real login. Three real connectors live (HubSpot, GA4, Google Ads). The IG playbook RAG retrieves chunks. Per-session cost tracking is accurate. p95 ask latency is under 12 seconds. The pilot's actual engagement data is mirrored in the app.
• Day 90. Maggie has been using the product for 30 days. You have shipped fixes from her usage daily. Two more pilots are signed. You and Chris have agreed on the v1.0 feature priority list.

What you should know about working with IG

We are a ten-person boutique consulting firm in San Jose. We have run client engagements in marketing, brand, growth, and operations for more than two decades. The core team is Chris (founder), David (operations), Mark (strategy and creative direction), Andy (SEO and restaurant vertical), and a rotating cast of contractors. The firm has revenue, profit, and a real client book. Next Best Action is a strategic project funded out of operations, not a bet-the-firm pivot.

Our editorial voice is direct, specific, and refuses fluff. We do not write "AI is changing everything." We write "here is what changed and here is what to do about it." Every doc you produce should sound like an IG doc.

You will work directly with Chris. You will demo to David, Mark, and Andy weekly. You will be on a daily fifteen-minute standup and a weekly sixty-minute leadership review. You will be on slacks-and-camera-on Zoom calls. You will be expected to push back when the product call is wrong.

We respect your time. Standups are fifteen minutes for a reason. We do not run meetings to fill time and we do not ask you to be on video for sake of it. We expect the same respect from you for our team's time.

How to apply

Email chris@innovativegroup.io with:

1. The 200 to 400 word response to the application question above.
2. Links to two products you have shipped end-to-end. Live URLs preferred. Private repo invites are fine.
3. A one-paragraph reaction to the architecture spec we share with you. Tell us what you would change.

Subject line: "Next Best Action engineer, [your name]"

We respond within three business days, no exceptions.

A note on the moment

We are a ten-person agency taking a real shot at productizing a piece of the work the Big Four cannot do at the price the mid-market will pay. The window for this is twelve to twenty-four months. We are not pretending this is a sure thing. We are also not pretending it is a small thing.

If you read this and your reaction is "yeah, that is the bet I want to be part of," we want to talk to you. If your reaction is "I want a more polished company with more headcount," that is a fair preference and we are not it.

Either answer is useful. The wrong move is to come in halfway and find out at week six that you are not in.

Apply by email →

Innovative Group is an equal opportunity employer. We welcome applicants from every background, nationality, and walk of life.