The setting
develops an AI answer engine that searches the web and replies with cited sources. is the version for large organizations, including regulated industries, with administrator controls over access and use.
A failed membership test does not establish whether provisioning or shared test state is wrong.
Repeated reproduction protects diagnosis, while repeated identity changes obscure the sequence being investigated.
Who you’d be doing this for
“I removed someone from the group, then spent the afternoon checking whether they'd actually lost access.”
Adanna Appiah · IT Administrator · a character in this brief
Manages employee provisioning for a regulated organization preparing a broader rollout.
What is at stake
Engineers lack an agreed reproduction for nine of 24 failed membership tests. You decide which identity changes AI may replay in isolation and which reproduction attempts need an engineer's approval.
Why it isn’t already fixed
Every obvious fix costs something else. That’s the part you’d have to decide.
- faithful replay vs clean test isolation
- failure frequency vs access consequences
- recent changes vs demonstrated causality
- richer context vs identity data minimization
Why Perplexity
WorkOS integrations and SCIM membership behavior underpin the access controls regulated administrators rely on.
Written with these in mind
Not your kind of problem? 48 more at Perplexity, or browse every organization.
Where the organization context comes fromThe postings this brief was written from, and the reporting read before it.
The organization context comes from public information. The situation and characters are written for the brief.
Job postings it was written from
Public descriptions of the work at Perplexity. They are listed as sources, not as openings. The lines under a posting are its own words, where it meets this brief.
- Member of Technical Staff (iOS Engineer, Computer Growth) (opens in a new tab)
- Build the flag-driven and server-driven patterns that let us test without waiting on a store release.
- Time spent at a fast-growing startup or on a high-ownership engineering team.
- Engineering Manager (Enterprise) (opens in a new tab)
- Enterprise Control Plane: Building the identity, security, and governance foundations that enable large organizations to deploy Perplexity confidently, including SSO, SCIM, role-based access, customer-managed encryption keys (BYOK), admin APIs, and enforceable policies governing access to data, models, and tools.
- Enterprise control plane: identity and access management, SSO/SCIM, role-based access, key management, admin APIs, policy enforcement, audit logging, data governance, or compliance and residency requirements for regulated industries.
- Member of Technical Staff (Software Engineer, Enterprise Trust & Governance) (opens in a new tab)
- Design and ship secure, well-documented admin APIs for audit logs, members, groups, roles, and settings.
- Own enterprise identity and provisioning, including SSO (SAML/OIDC), SCIM, WorkOS integrations, and support for identity providers such as Okta and Microsoft Entra ID.
- Member of Technical Staff (Senior QA Web Engineer) (opens in a new tab)
- Cursor, Claude Code) to navigate the codebase, debug issues, and ship small fixes independently
- Fix small, low-risk issues yourself where appropriate, shipping simple fixes to reduce engineering overhead and speed up resolution.
- Member of Technical Staff (GTM Engineer, Growth) (opens in a new tab)
- Alongside product engineers, you ship production code, instrument the product and run experiments.
Reporting read before this brief was written
The pieces closest to the situation in this brief.
Perplexity published its agent-safety design on September 29, 2026, setting a rule that at least one safeguard layer must be deterministic code outside the agent's control, and listing real-time audit logs for Computer and Comet agents among its deployed layers.
How we engineer safer agents (opens in a new tab) · perplexity.ai · Sep 29, 2026
This is the setup. The work is inside.
Running it puts you in the room: the full situation and its constraints, stakeholders who push back in their own words, and the decisions that are yours to make. What you produce becomes a Day One Plan — work you can show someone instead of describing.
Where similar work appears
Other live briefs connected to this one, and what connects them.
Similar problems at other organizations
Like this one, each of these is about people starting something and never quite finishing it.
Security engineers must judge 240 production grants, including rarely used recovery access.
Where will you require an owner’s explanation before AI can propose revoking a quiet grant?
Administrators cannot quickly explain why access changed after a policy update, even when records exist.
You must decide what evidence AI can rank and what engineers must confirm before an answer leaves the team.
Administrators repeat policy changes when expected notifications do not arrive.
You must decide where the AI's trace of a failure stops and the on-call engineer's confirmation of the cause begins.
The same setting, from another role
Each is set around Perplexity’s Perplexity Enterprise, written for a different role. They’re separate briefs, not the same scenario.
Regulated IT administrators wait while agreement requests arrive without key context.
You must decide whether the system only proposes routes or also holds incomplete matters for counsel.
Regulated IT administrators cannot yet govern Computer access by role.
You must decide which controls belong in reusable terms and which require customer-specific approval.
A healthcare IT administrator cannot enable Computer until the BAA commitments are settled.
You must choose which agent commitments belong in the signed terms and which cannot be promised yet.
Other problems with a similar purpose
In each of these, as in this one, the aim is to engineer security-critical controls whose correctness holds up even when carefully examined by skeptics.
Held content can be classified in most samples, but nothing proves that every retrieval surface enforces the split.
You have to ship enforcement, define the proof, and get sign-off before the irreversible migration.
A controller trust-path concern must be addressed within seven days across occupied sites.
You must weigh stronger remote protection against the chance of disrupting reliable door access.
An automated grader for AI-generated Python passes 97% of tests, but reviewers found real errors in code it marked safe.
You have to decide what evidence clears this for launch, since the code runs against customer warehouse data.
