Senior AI Engineer

About Discovered LabsAt Discovered Labs we work with $10M - $50M ARR companies to help them get more leads, users and customers from Google, Bing and AI assistants such as ChatGPT, Claude and Perplexity.We approach marketing the way engineers approach systems: data in, insights out, feedback loops everywhere. Every decision traces back to measurable outcomes. Every workflow is designed to eliminate manual bottlenecks and compound over time.High-level overview of our approach:Data-driven automation: We treat marketing programs like products. We instrument everything, automate the repetitive, and focus human effort on high-leverage problems.First principles thinking: We don't copy what others do. We understand the underlying mechanics of how search and AI systems work, then build solutions from that foundation.Full-stack ownership: SEO and AEO rarely work as isolated tasks. We work across the entire funnel and multiple surface areas to ensure we own the outcome and clients win.The Agency OSWe're building the first agency OS: a system where the agency's work is the software, enabling clients to gain an unfair advantage.As Karpathy puts it, Software 3.0 automates what humans can verify. Verifiability, not capability, decides which work can safely be handed over. So every eval we write expands the set of work we can trust an agent with, and the harness isn't overhead around the product, it's what makes the product possible. Dex Horthy covers the operational half: the agents that survive contact with customers are well-engineered software with human judgement placed deliberately, and disciplined loops beat vibes.What that looks like here: expert orchestrators, not operators. One strategist supervising a fleet of agents running real client workflows at scale, with the system surfacing what genuinely needs a human decision and staying quiet about what doesn't. Feedback loops everywhere. Every correction an expert makes becomes an eval case, and the swarm is better the next cycle.The TeamYou'll join the Automations team at Discovered Labs. We build the agent swarm: a growing fleet of agents and workflows that do real client work end to end, and the engineering that makes them trustworthy enough to let loose on it.Your mandate is both halves of that. Expand what the swarm can do, and prove that what it does is right. That means applying hardcore engineering to a famously slippery problem: making non-deterministic systems reliable, measurable and provable. Evals that fail a build. Traces that explain a decision. Guardrails that hold when a model doesn't. Agents that know what they don't know and say so.You'll work alongside the Automations lead, and with our AI & Data team who own the platform and tooling underneath you. You won't be building orchestration from scratch or fighting for infrastructure. It's there, and it's good. Your job is to take it the last mile.Every agent you make provable is an agent we can put in front of a client. That's the whole game, and the ceiling on how fast this company grows.We're a deeply technical team building the SpaceX of the AEO & SEO space. You'll work alongside engineers who have built fraud engines powering Stripe, shipped AI code review at CodeRabbit, built at Amazon, developed self-driving car systems at Aurora, and conducted AI research at Stanford. We don't have layers of management. You'll work directly with founders who can go deep on architecture, code, and product.This RoleYou'll join the Automations team, and your job is to productionise what we've built.We have live 24/7 agents, data pipelines and workflow infrastructure. What you’ll be doing is helping us get the stack running robustly, reliably and at scale.That means evals that catch regressions before they ship, observability that explains why an agent did what it did, provenance that traces every claim back to its source, and interfaces that let a non-engineer SEO strategist run and review the work. You'll build the harnesses and eval suites that gate agent changes in CI, the lineage and step-trace tooling that makes agent reasoning inspectable, and the surfaces the team actually uses. Then you'll close the loop: instrument the output, feed the signal back, and make the agents measurably better every cycle.You'll also extend what agents can reach. Today they work over APIs and our own data. Next they operate real websites along withsession handling, and navigating interfaces that were never designed for machines.The hard problem is reliability and legibility. An agent that's right 80% of the time and can't tell you which 80% is worthless. An agent that's right 95% of the time, shows its working, and flags its own uncertainty is a product.You report to the CTO and work alongside the Automations lead, with our AI & Data team owning the platform beneath you. You own your evals, your CI, and your monitoring.What You'll DoAgent eval harnesses. Golden suites, deterministic checks, and online scoring that gate every agen