Chris Shayan¶
Head of AI at backbase. Practitioner building agentic systems for retail / digital banking. Public-facing via LinkedIn essays on the gap between agent demos and production.
Position (May 2026)¶
In a widely-shared LinkedIn post he articulated the three walls every banking agent team hits in production — signal vs noise, state & memory, consequence modeling — and proposed a three-component architecture (Signal Catalogue, Digital Twin, Nudge Mesh) sitting around the LLM rather than inside it. His framing — "the LLM is essential but insufficient… the linguistic engine… but it needs architecture around it" — is a domain-specific restatement of the 2026 consensus captured in harness-engineering and control-flow-vs-prompt-flow.
See the full cross-reference in chris-shayan-backbase-intelligence-layer-2026.
Why he matters to this wiki¶
First non-dev-tooling, non-ML-platform practitioner in the wiki to converge on the harness-centric view. Banking has different constraints (regulated consequences, silent-customer drop-off is the default feedback signal, audit trails matter legally) so his independent arrival at the same architecture is cross-domain evidence for the thesis, not another voice inside the echo chamber.
Cross-references¶
- backbase — employer; the Intelligence Layer is their product.
- chris-shayan-backbase-intelligence-layer-2026 — the synthesis query filing his post against the wiki.
- harness-engineering · learning-agent-loop · silent-failure-dropoff — the three concepts his three "walls" map to.