Toast Interview — 1-Hour Revision Guide

Vivek Jain · Senior Principal Engineer, AI Team Agent · Hiring Manager: Darien Ford

⏱ How to use this guide

| Minutes | What to do | |—|—| | 0–15 | Read Section 1 — Who you are & your 5 STAR stories | | 15–30 | Read Section 2 — Toast domain & your agent vision | | 30–45 | Read Section 3 — All 12 questions + your answers | | 45–55 | Read Section 4 — Gaps, bridges & questions for Darien | | 55–60 | Read Section 5 — BrightHire tips & opening 90 seconds |

Section 1 — Who You Are (your 3-sentence pitch)

“I’m a Principal Software Engineer with 18 years building distributed systems and data platforms. For the last 5 years at JPMC I’ve been building agentic AI infrastructure — a metadata-driven RAG platform that lets AI agents discover datasets, generate governed SQL, and execute auditable actions across enterprise data. I want to bring that architecture to Toast’s workforce domain — where the stakes are real for restaurant operators and their staff.”

Your 5 STAR Stories (know these cold)

Story 1 — The Federated Platform (Architecture Vision)

Use for: Q2 complex system, Q9 exec influence, Q5 product roadmap

Story 2 — Agentic AI + OPA (Governance Story)

Use for: Q1 agent vision, Q3 coding agents, Q6 year one priorities

Story 3 — Product Roadmap Influence

Use for: Q5 product influence, Q9 exec communication

Story 4 — Mentoring a Senior Engineer

Use for: Q7 mentoring

Story 5 — MCP Agent Architecture (Your New Story)

Use for: Q3 coding agents — YOUR STRONGEST NEW ANSWER

Section 2 — Toast Domain & Your Agent Vision

What Toast’s Team product does (say this naturally)

“Toast’s Team area is essentially a real-time financial compliance system. Every clock-in, shift edit, and tip allocation has downstream tax and wage-law consequences. Labor is 30–35% of a restaurant’s revenue — it’s the most controllable cost and the most complex to manage.”

Your Year One Agent Priorities (Q6 — know this cold)

Priority 1 — Overtime risk detection

“Highest ROI. Labor is 30–35% of revenue. An agent that flags overtime risk before it happens and suggests coverage options has immediate, measurable value.”

Priority 2 — Tip pool audit trail

“Tip discrepancies destroy staff trust. An agent that explains every cent of a tip calculation in plain English removes the #1 source of manager-staff friction.”

Priority 3 — Multi-state compliance monitoring

“Restaurant groups operate across states with different tip credit laws and overtime rules. An agent that surfaces compliance gaps before payroll runs protects operators from costly violations.”

Landing line: “All three are trust-building agents, not efficiency agents. When an operator trusts the agent’s outputs, everything else follows.”

Your Agent Architecture for Toast

Toast POS (live events: clock-ins, tips, orders)
        │
        ▼
Toast Metadata MCP          Toast Query MCP
(employee roles,    ◄──────► (hours worked,
 tip policies,               tip pool calc,
 wage rules,                 overtime scan,
 location config)            audit log)
        │                         │
        └──────────┬──────────────┘
                   ▼
         Governance / Policy Layer
         (labor law rules, FLSA,
          audit trail, human approval
          gate for write actions)
                   │
                   ▼
         Toast Workforce Agent
         (Claude claude-sonnet-4-20250514)

Say this: “I actually built this — a working agent with two MCP servers. Toast Metadata MCP mirrors OpenMetadata from my JPMC work. Toast Query MCP mirrors the Trino layer. The governance layer sits between the agent and the MCP servers, so every action is policy-checked before it runs.”

Key Toast facts (drop these naturally)

• 156,000 restaurant locations
• $2B+ ARR, first GAAP profitability in 2024
• ~11% penetrated in total addressable market
• Toast Pay Cards — instant tip access after shift close
• “No Tax on Tips” — H.R.1 (2025–2028) directly affects Toast’s payroll product
• POS = Point of Sale = the tablet/terminal where orders are taken, payments processed, tips given, employees clock in/out

Section 3 — The 12 Questions

Q1 — Architectural vision for AI agents in workforce management

(Most important question — Darien’s #1 probe)

Your answer structure:

1. Frame the domain: “The Team area is a real-time financial compliance system…”
2. Describe the architecture: metadata layer → reasoning layer → governed action layer
3. Name the first 3 agents (overtime, tip audit, compliance)
4. Bridge to JPMC: “Same pattern I built at JPMC — the domain changes, the architecture doesn’t”

Q2 — Most complex distributed system you’ve designed

→ Use Story 1 (Federated Platform) Land on: “200K queries/day, 99.9% reliability, weeks to hours”

Q3 — Hands-on experience with coding agents

(Highest-risk question — your MCP story saves you)

Your answer:

“My core experience is building the agent infrastructure layer at JPMC — the metadata discovery, governed tool execution, and OPA policy enforcement that enables agents to operate safely. Concretely: I connected an agent to OpenMetadata MCP and Trino MCP — so the agent discovers datasets, understands their structure, generates queries, and executes them with a policy gate on every action. I’ve also built the Toast equivalent using the Anthropic SDK — a workforce agent with two MCP servers and a governance layer. I see spec-driven development as the right discipline for production agents — vibe coding gets you a prototype, SDD gets you something you can operate.”

Q4 — How to spread agentic patterns across Toast

Three phases:

1. Establish — Build Team agent as the reference implementation with constitution.md
2. Enable — Reusable scaffolding library + shared policy layer + working group (1 engineer per FinTech team)
3. Measure — Adoption velocity tracked publicly, executive visibility

Bridge: “I did this at JPMC — federated platform went from one team to 1,000+ users across multiple business units. Same playbook.”

Q5 — How you influenced a product roadmap

→ Use Story 3 (Product Roadmap) Land on: “Working code is the most powerful product influence tool.”

Q6 — Year one priorities for Toast Team agents

→ See Section 2 priorities above Land on: “Trust-building agents, not efficiency agents.”

Q7 — How you mentor senior engineers

→ Use Story 4 (Mentoring) Land on: “Best mentoring isn’t teaching — it’s creating conditions where someone realises they were capable all along.”

Q8 — A time a system failed at scale

Your structure:

• What system, what scale, what broke
• How you found out (metrics? alert? angry call?)
• Immediate containment steps
• Root cause analysis
• Systematic fix (not just a patch)
• End with: “The thing that changed in how I design systems after this is [X]. I’ve applied it ever since.”

⚠️ Prepare your own real story here. Don’t make one up — Darien was CTO, he’ll probe.

Q9 — Executive communication and influence

Your answer:

“Executives decide on risk, cost, competitive position and time — not technical merit. I translate technical options into those terms first. When I proposed the agentic platform at JPMC, I didn’t lead with RAG architecture. I led with: ‘Today it takes a security analyst a week to answer a threat question that should take 4 seconds. Here’s what that costs us.’ The architecture came after the business case landed.”

Q10 — Why Toast, why this role

Three parts:

1. Toast — $2B ARR, 156K locations, 11% TAM penetration. Inflection point where agents compound the value of every product.
2. Role — Workforce management in hospitality is genuinely hard. Real-time financial compliance for high-stakes operator decisions.
3. Fit — “My entire 5 years at JPMC has been preparing me for this: governed, trustworthy agent infrastructure for high-stakes decisions. I want to apply that at the product layer, not just infrastructure.”

Q11 — SaaS gap: enterprise platforms vs customer-facing product

Your bridge:

“You’re right I haven’t owned a customer-facing SaaS product. What I’ve run is a platform 1,000+ engineers depended on daily with contractual SLAs, incident escalation, a deprecation process, and versioned APIs — the operational complexity maps directly. What I’d invest in immediately is time with restaurant operators. Not to learn the basics from a PM — to sit with a manager during a shift change and watch what they actually do. The best architectural decisions I’ve made came from watching real users.”

Q12 — Your questions for Darien

1. “What does success look like for the first Team agent in production — from an operator’s perspective, not just a technical one?”
2. “How does the feedback loop work between restaurant operators and this team? What signal tells you an agent is trusted vs just used?”
3. “You’ve transformed engineering orgs at Capital One and Madhive — what’s the cultural shift you’re driving at Toast with agentic development, and where’s the resistance?”
4. “What does this role need to accomplish in the first 90 days to make you say the hire was the right decision?”

Section 4 — Gaps & How to Bridge Them

About Darien Ford

• Former CTO at Madhive, Sr. Director Engineering at Capital One, Sr. Director at Knighted
• Loves building products and teams — not just systems
• Heavy AWS background (Capital One)
• Will probe: Can you be the most senior technical voice in his LoB? Do you have genuine passion for agents? Will you shape product, not just execute?
• What impresses him: Business outcomes first, technical detail second. Working code > slide decks. Honesty about gaps + a plan to close them.

Section 5 — Final Prep

Your opening 90 seconds (practise this out loud)

“I’m Vivek — Principal Software Engineer, 18 years building distributed systems and data platforms. The last 5 years at JPMC have been spent almost entirely on agentic AI infrastructure: I designed a metadata-driven platform where AI agents discover datasets, generate governed SQL queries, and execute auditable actions across enterprise data — serving 1,000+ users at 200,000 queries a day. I connected that agent to OpenMetadata and Trino through MCP servers — same two-layer pattern I’d want to apply to Toast’s workforce domain. I’ve spent the last week building the Toast version as a prototype. I’m excited to talk about it.”

BrightHire — 5 rules

1. Use STAR explicitly — Situation → Task → Action → Result. Every answer.
2. Pause 2 seconds before answering — cleaner transcript, better scoring
3. Say numbers out loud — 200K queries, 99.9%, 1,000 users, weeks → hours
4. Mirror the JD language — “agentic”, “spec-driven development”, “mission-critical”, “FinTech product strategy”, “governed”
5. Don’t fill silence — Darien taking notes on BrightHire looks like silence. Wait for the next question.

Spec-Driven Development (SDD) — one paragraph

“SDD means writing the specification before writing any code — defining what to build, the constraints, the acceptance criteria, and the non-negotiables. The spec is the primary artifact; code is the output. For Toast’s Team agents I’d start with a constitution.md encoding the non-negotiables: FLSA compliance checks, tip calculation audit requirements, payroll accuracy guarantees. Agents implement features within those rails. Vibe coding gets you a prototype. SDD gets you something you can operate, audit, and maintain.”

Key vocabulary to use naturally

One-liners to land key points

• On the domain: “This isn’t a chatbot problem — it’s a payroll compliance problem that happens to use AI.”
• On your architecture: “The pattern is the same — metadata-aware agents with governed execution. The domain changes, the architecture doesn’t.”
• On MCP: “MCP is the right abstraction — build the governed data layer once, every agent plugs in.”
• On SDD: “Vibe coding gets you a prototype. Spec-driven development gets you something you can operate.”
• On product influence: “Working code is the most powerful product influence tool — more than any slide deck.”
• On trust: “The first agents I’d ship are trust-building agents, not efficiency agents. Trust unlocks everything else.”

Quick Reference Card

JPMC → Toast mapping
─────────────────────────────────────────────────
OpenMetadata MCP   →   Toast Workforce Metadata MCP
Trino MCP          →   Toast Query / POS MCP
Cybersecurity data →   Workforce data (hours, tips, shifts)
OPA + MOAT         →   Labor law + payroll compliance gates
200K queries/day   →   156,000 restaurant locations
Weeks → hours      →   Manager decisions: minutes not days

Your 5 stories
─────────────────────────────────────────────────
1. Federated platform     → Architecture, scale, reliability
2. Agentic AI + OPA       → Governance, agent infrastructure
3. Product roadmap        → Influence, exec communication
4. Mentoring engineer     → People, leadership
5. MCP agent (new!)       → Coding agents, hands-on proof

Toast numbers to drop
─────────────────────────────────────────────────
156,000 locations · $2B+ ARR · 11% TAM penetration
30-35% of revenue = labor · <10 min payroll run

Good luck Vivek. You’ve built the architecture. You understand the domain. You have the stories. Go show Darien what you’d build.