Building a Neobank With Agentic Engineering
summary
Natech reshapes product and engineering around agents and moves specs and implementation plans in the repository (GitHub issues and MD files) so regulatory decisions are auditable; routine coding is delegated to agents while humans review GDPR, security and risk-sensitive changes under a review plans before code rule. The system uses five layers of verification (specs, static checks, dynamic tests, agent review, human review) and produces the Snappi mobile bank — built Snappi in under a year and live with ~200k clients.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 1:08Natech and the Snappi neobank
- 2:11The scope of the talk
- 3:00A process built for an uncertain launch date
- 3:58Turning requirements into a delivery loop
- 4:35Starting from the existing operating model
- 5:13Choosing a source of truth
- 5:55Specifications and plans in the repository
- 6:14More fluid roles across the team
- 6:39Product ownership and AI-assisted specification
- 7:03Design systems and generated user flows
- 8:11Taste, accountability, and ownership
- 9:13The review debate
- 9:54Reviewing plans before code
- 11:14Security and compliance requirements
- 11:55Five layers of verification
- 12:28Trusting agent-generated code
- 14:02Redefining engineering seniority
- 15:33Engineers as managers of agents
- 17:14Avoiding cognitive surrender
- 18:38The new operating model
- 19:27Building Snappi in under a year
- 20:10Closing
related talks
33mOptimizing Bank Operations: Design Patterns, NFR, and GenAI Fusion
01h 04mPanel Discussion: Money Moves at Machine Speed - Trust, Risk, and Quality in Agentic Finance
45mAI Tools in Regulated Banking: Monzo's Control Model
49mPanel Discussion: Driving the Agentic Shift in Banking Adoption Governance and Scale
19mBuilding Secure Services with AI Agents: Practical Patterns
33mAgent Experience
01h 04mSpec-Driven Development: Agentic Coding at FAANG Scale and Quality
21mThe Agent Native Company