Skip to content

Mikhail Parakhin

CTO of shopify (2024–). Previously CEO of one of microsoft's business units — the Web & Ads group encompassing Bing, Edge, MSN, and the team that shipped Sydney (the early bing/GPT-4 chatbot whose emergent personality became the canonical "AI with a soul" story).

On Jensen's "$500k engineer, $5k in tokens" line

Agrees Jensen Huang is directionally correct that engineers under-consume tokens, but says raw token count is the wrong metric. What actually matters: a strong critique loop — one agent writes, a second (ideally a different model) critiques; latency goes up, quality goes up. The anti-pattern is running many parallel non-communicating agents — "almost useless."

"It's not about just consuming tokens. […] The anti-pattern is running multiple agents too many agents in parallel that don't communicate with each other. That's almost useless."

Connects to do-not-outsource-thinking and ai-generated-code-is-untrusted — volume without review multiplies bugs into production, even if the model's per-line bug rate is below a human's.

At Shopify

Drives Shopify's internal AI stack. Four named systems he owns or sponsors: tangle (reproducible ML workflows), tangent (auto-research on top of Tangle), simgym (customer simulation), and Shopify's adoption of liquid-ai models for latency- and context-bound jobs. Public face of Shopify's shopify-ai-adoption-inflection (approaching 100% daily-active AI usage after a December 2025 model quality step change).

Sydney backstory

The most interesting anecdote he shares: Sydney was first shipped in India and nobody noticed for weeks. The initial implementation didn't use an OpenAI model — it ran on a microsoft×NVIDIA Megatron collaboration model. The emergent personality was not accidental; personality shaping was deliberate.

See also