AI Summary
→ WHAT IT COVERS Woodson Martin, CEO of OutSystems, explains how the enterprise software platform serving Toyota, Vodafone, and Petrobras maintains its "software that never breaks" promise while accelerating AI-driven development. OutSystems went from 4 major feature releases in Q4 2024 to 26 in Q2 2025, using abstraction layers, model routing, and agentic tooling to govern enterprise-grade AI delivery. → KEY INSIGHTS - **Abstraction Layer as AI Guardrail:** Rather than letting coding agents manipulate code directly, OutSystems has agents manipulate an abstract application model, then deterministically generates compliant code. This means every AI-built asset automatically inherits enterprise security controls, GDPR/HIPAA compliance, and role-based access controls without requiring developers to configure them manually — eliminating the core reliability problem of vibe-coded enterprise software. - **Token Spend Optimization via LLM Routing:** OutSystems reduced token costs below forecast by building a custom gateway that routes tasks to lower-cost models based on complexity. Peak spend hit in June–July 2025, then declined after implementing the router. The core finding: most enterprise workloads — PDF parsing, data structuring, workflow automation — run adequately on older or smaller models, not frontier ones like Opus. - **Velocity Benchmark from AI-First Engineering:** OutSystems shipped 4 major features in Q4 2024, 19 in Q1 2025, and 26 in Q2 2025 after adopting an AI-first internal engineering model. The acceleration came from combining frontier models with a custom internal harness loaded with company-specific context. Organizations can replicate this by building context-rich internal harnesses rather than relying on generic coding agents alone. - **Agent Foundry for ROI-Driven Agentic Prioritization:** OutSystems built an internal "agent foundry" that analyzes telemetry from existing customer applications, identifies manual processes suitable for automation, forecasts time-savings ROI, and generates a V1 agentic system with one click. This directly addresses the hardest enterprise AI challenge: knowing what to build first. Organizations should instrument existing apps to surface automation opportunities rather than guessing. - **Compliance Backlog as the Real Deployment Bottleneck:** Many enterprise agentic systems are fully built and tested but stuck in compliance queues awaiting approval of specific AI models — including scrutiny over whether training data was legally acquired. Organizations should engage compliance and legal teams during the design phase, not post-build, and prioritize lower-risk workflows like IT resource allocation before tackling regulated decision-making processes like loan origination. - **Platform Trust Compounds Over Time and Cannot Be Replicated Quickly:** Enterprise readiness is not just uptime or security features — it is the accumulated track record of mission-critical deployments across regulated industries. Regulated enterprises in Europe and Asia evaluate AI platforms partly on years of proven delivery history. Startups with superior technology still face this trust deficit, meaning incumbents with deep deployment records hold a durable structural advantage in regulated verticals. → NOTABLE MOMENT OutSystems demonstrated a banking use case where a consumer chatting with Claude about a home renovation was seamlessly identified as pre-approved for a loan, transitioned into the bank's mobile app mid-conversation, and processed through back-end agents — with no redundant document requests — all while meeting regulatory requirements. 💼 SPONSORS [{"name": "Parallel", "url": "https://parallel.ai/tcr"}, {"name": "Anthropic (Claude)", "url": "https://claude.ai/tcr"}, {"name": "OutSystems", "url": "https://outsystems.com/tcr"}, {"name": "Tasklet", "url": "https://tasklet.ai"}] 🏷️ Enterprise AI Governance, LLM Cost Optimization, Agentic Systems, Low-Code Platforms, Software Compliance, AI-Driven Development Velocity