Staff Engineer, Applied AI

Match Group, Palo Alto, California +2

What you will do

Work closely with internal teams to build real solutions for challenges they are facing day to day.
Own the design of the agent platform across gateway, connectors, identity, policy, guardrails, observability and cost, and write the roadmap for how it evolves.
Design identity and delegation for agents: acting on behalf of a user, agent-owned identities for unattended jobs, short-lived scoped credentials, and every action traceable to a responsible human.
Build access control beyond tool allowlists: per-action, per-argument and per-data-class policy, human approval for irreversible actions, and policy that's reviewable and tested.
Build guardrails enforced at the tool boundary, not in prompts, against prompt injection through tool output, confused-deputy access, data exfiltration and over-broad scopes.

Match Group, Palo Alto, California +2

$225K–$265K

a year

This base pay is below the median for staff ML and AI engineering jobs in the Bay Area.

We compare the middle of each posted range. This is base pay as posted. The group does not include this job.

The Match Group site opens in a new tab.

No one has reported an interview at Match Group yet. Did you interview there? Add the questions.

Compare your résumé with this job. See where you match, what is missing and which lines to change.

Work closely with internal teams to build real solutions for challenges they are facing day to day.
Own the design of the agent platform across gateway, connectors, identity, policy, guardrails, observability and cost, and write the roadmap for how it evolves.
Design identity and delegation for agents: acting on behalf of a user, agent-owned identities for unattended jobs, short-lived scoped credentials, and every action traceable to a responsible human.
Build access control beyond tool allowlists: per-action, per-argument and per-data-class policy, human approval for irreversible actions, and policy that's reviewable and tested.
Build guardrails enforced at the tool boundary, not in prompts, against prompt injection through tool output, confused-deputy access, data exfiltration and over-broad scopes.