Founding AI Scientist (Agentic Systems) – Glitch Ads

Founding AI Scientist (Agentic Systems)

By Dave Kearney • 5 min read

At Glitch, we’re reimagining how B2B companies grow. Our AI-driven platform helps marketing teams unlock smarter, faster lead generation across major ad platforms, transforming how businesses discover and engage their ideal customers.

We believe the future of growth is orchestrated by agentic systems with tool use, memory, and continuous learning, not dashboards and manual knobs. GlitchAds.ai is not an ad network or a rules engine. It is an autonomous AI platform where agents plan, act, and learn across marketing workflows, spinning up campaigns, generating creative, pacing budgets, and self-correcting in live traffic.

We’ve recently raised our seed round and are now expanding rapidly across the US and Europe. This is a rare opportunity to define and build the scientific and architectural foundations of an AI-first company where autonomy, safety, and scale come together.

The Opportunity

As the Founding Lead AI Scientist, you will architect and deliver the next generation of agentic AI systems for marketing automation. You will work at the core of Glitch’s intelligence layer, designing, training, and deploying autonomous agents that make and execute growth decisions in real time.

You will operate across research, engineering, and product boundaries, taking models from concept to production, establishing experimentation frameworks, and translating cutting-edge AI into tangible platform capabilities.

Over the next 12 months, your focus will be on:

  • Designing and deploying agentic architectures that plan, act, and learn in complex environments
    Implementing continuous evaluation, safety, and learning loops for production-grade autonomy
  • Defining the technical roadmap for Glitch’s agentic control systems and AI infrastructure
  • Establishing research and experimentation practices that drive measurable business outcomes

Areas of Responsibility

Agentic Architecture

  • Design multi-agent and hierarchical planners (planner, executor, critic) with tool use, working memory, retrieval, and self-reflection loops (for example, ReAct or Reflexion paradigms for production)
  • Build agents capable of tool-augmented execution across APIs, bidding, creative generation, analytics, and constraint solving
  • Enable agents to verify outputs, recover from failure, and operate safely under uncertainty

Learning and Control Systems

  • Develop contextual bandits and reinforcement learning for bidding and budget allocation
  • Implement online learning with safety envelopes and reward shaping aligned to business KPIs such as CPA, ROAS, and LTV
  • Build autonomy guardrails including policy constraints, cost caps, anomaly detection, and human-in-the-loop controls

Evaluation and Simulation

  • Create offline, shadow, and canary pipelines for iterative agent testing and deployment
  • Develop agent simulators, evaluation harnesses, and red-teaming frameworks to ensure robustness, drift detection, and resilience to adversarial inputs

LLM Systems

  • Integrate retrieval-augmented generation for keyword and creative workflows
  • Apply program synthesis and prompt orchestration to manage complex tool plans
  • Optimize latency and cost with adapter, distillation, and caching strategies

Science to Product

  • Work cross-functionally with product and design to frame objectives, define metrics, and run rigorous experiments
  • Translate research breakthroughs into production-grade capabilities that customers can toggle, audit, and trust.

Key Outcomes

  • An agentic loop in production that outperforms baseline bidding and pacing models with measurable gains in CPA, ROAS, and time-to-launch
  • A hardened evaluation and safety harness with datasets, metrics, guardrails, and rollback protocols
  • An LLM and tools pipeline that reduces setup time and automates negative keyword curation with traceable reasoning
  • Clear documentation, experiment logs, and a hiring plan for the next AI and ML roles

Experience and Skills

Must-Haves

  • 5–7 years of applied ML experience with shipped systems that impacted real-world KPIs
  • Expert-level Python and deep hands-on experience with PyTorch across training, serving, and online iteration
  • Proven experience in at least two of the following areas:
    • Agentic LLM systems (tool use, planning, memory)
    • Reinforcement learning or bandits
    • Retrieval and ranking
    • Time-series or anomaly detection
    • Causal inference or experimentation
  • Strong background in statistics, experiment design, and counterfactual reasoning
  • Demonstrated ability to build safe, autonomous systems with measurable business alignment

Nice-to-Haves

  • Experience with ads, growth platforms, or control systems under constraints (RTB, pricing, marketplaces)
  • Familiarity with distributed systems such as Ray, Triton, Kafka, feature stores, or vector databases
    Experience with FastAPI, async Python, Kubernetes (GCP or AWS), SQL, and modern data tools such as DuckDB, Polars, or Pandas
  • Publications or OSS contributions in agentic AI, reinforcement learning, or NLP where impact matters more than pedigree

What Makes You Stand Out

  • You love the intersection of autonomy and safety and you ship agents that act, verify, and recover
  • You move fluidly from notebook to service, valuing “simple that wins” until cutting-edge is the simplest thing that wins
  • You can define rewards and objectives clearly and defend them with data
  • You are comfortable owning architecture while writing production code daily
  • You are excited by the challenge of taking research ideas to scalable, reliable systems

What We Offer

  • A defining role in shaping the core intelligence of Glitch’s agentic platform
  • Direct collaboration with the founders on AI, product, and strategy
  • Competitive compensation with meaningful equity
  • Flexible, remote-friendly environment with US and EU time zone overlap
  • A culture built on trust, experimentation, and technical excellence

And More as We Grow

We are still early, which means you will have the autonomy to shape not just models and systems but how the company itself approaches AI. If you are driven by the challenge of building self-learning, self-correcting systems, this is your chance to define the future of agentic growth