Internal agentic workflow platform pairing deterministic code with restricted LLM steps
A platform where repetitive engineering tasks are defined as YAML workflows that live in a repo and go through normal merge request review. Deterministic code handles the predictable steps; an LLM step is called only where judgment is needed, with read-only permissions, strict JSON output schemas and per-step restrictions on network, tools, skills and MCP servers. The model can read code but never modify it, and human merge-request review stays in the loop.
online BOTH AI Agents & Automation
From AI Engineer — Agents in Orbit: Workflows, Not Autopilot — Eduardo Simon & Javier Casal, Flywire at 03:04
Problem: Open-ended autonomous agents are unreliable and unsafe in a compliance-heavy environment; engineers still lose time to repetitive tasks like code review, translations, dependency fixes, ticket triage and bulk repository migrations. Structure keeps execution predictable, auditable and inside the company's own data boundary.
For: Platform/infrastructure teams in mid-size to large engineering organizations, especially in regulated industries (payments, finance, healthcare) handling PCI/PII/PHI data.
Products from this video
AWS ECS Fargate Claude Agent SDK GitLab Jira OpenTelemetry Orbit Sentry
Examples
- Automated code review agent: over 9,700 runs in 30 days, averaging under three minutes per merge request.
Other takes on AI agent platforms and compute access
- Autonomous AI agents that perform administrative labor for underserved small businesses, starting with dental practices
- A personal, open-source AI agent that can be operated through messaging channels such as WhatsApp and Discord
- AI virtual staging service for real estate listings
- Enterprise AI concierge for customer, sales, and operational workflows
- AI appointment-scheduling assistant for healthcare providers
- Autonomous SEC-filing research desk with self-graded scorecards