Design an AI agent platform: system design interview question

How the interviewer opens it

“Design a platform where AI agents complete tasks through multiple steps and external tools. The platform must enforce permissions, limit execution costs, and recover safely from failures.”

The prompt is vague on purpose. The first minutes are for you. Reduce the prompt to the core features and agree on the numbers before you draw.

What to clarify first

Scale
Run behavior
Tools
Permissions and approval
Costs
Latency
Read and write mix
Consistency and durability
History and secrets
Scope limits

The interviewer has the answers. In a SysRound round, Alex gives these answers, as a real interviewer does.

Scope of a strong answer

Core requirements

Start a run and execute model and tool calls across multiple steps
Enforce user permissions and require approval for sensitive actions
Limit costs and stop runs
Recover run progress without unsafe repeated actions

Usually out of scope

Train models
Execute arbitrary user code
Process financial transfers
Operate active-active regions

What interviewers ask about

Run progress and tool results: A tool can change an external system before the worker records success. A retry can repeat the change.
Who can use a tool: The model can request actions from untrusted text. User permissions can change while a run waits.
Execution cost and stop rules: Models can repeat actions or produce more text than expected. Retries and concurrent runs compete for the same budget.

Common mistakes

Treats an agent as one long request with no durable progress.
Assumes that a timeout means the tool did nothing.
Claims exactly-once external effects without provider support.
Gives the model responsibility for permission checks.
Checks costs after calls or resets the run budget at midnight.
Releases unknown charges or repeats successful actions during recovery.

The bar by level

Senior (L5/E5): Defines recovery states and provider-specific retry rules. Enforces atomic cumulative run budgets and daily tenant budgets. Handles approval changes, revoked access, cancellation, and stale workers.
Staff (L6+/E6+): Defines precise admission and external-effect guarantees. Explains where provider behavior limits those guarantees. Balances safe pauses against progress during failures and unknown charges. Extends budget and authority rules across concurrency, midnight, and regions.

Sources · 1 reference

These records explain where the question appears. A company tag or candidate report does not guarantee a future interview question.

SysRound · 1 record

No company association
Design an AI agent platform
Reported date not specified
Source snapshot: Nov 12, 2026
View source

How the interviewer opens it

“Design a platform where AI agents complete tasks through multiple steps and external tools. The platform must enforce permissions, limit execution costs, and recover safely from failures.”

The prompt is vague on purpose. The first minutes are for you. Reduce the prompt to the core features and agree on the numbers before you draw.

Scope of a strong answer

Core requirements

Start a run and execute model and tool calls across multiple steps
Enforce user permissions and require approval for sensitive actions
Limit costs and stop runs
Recover run progress without unsafe repeated actions

Usually out of scope

Train models
Execute arbitrary user code
Process financial transfers
Operate active-active regions

What interviewers ask about

Run progress and tool results

A tool can change an external system before the worker records success. A retry can repeat the change.

Who can use a tool

The model can request actions from untrusted text. User permissions can change while a run waits.

Execution cost and stop rules

Models can repeat actions or produce more text than expected. Retries and concurrent runs compete for the same budget.

Common mistakes

Treats an agent as one long request with no durable progress.

Assumes that a timeout means the tool did nothing.

Claims exactly-once external effects without provider support.

Gives the model responsibility for permission checks.

Checks costs after calls or resets the run budget at midnight.

Releases unknown charges or repeats successful actions during recovery.

The bar by level

Senior (L5/E5)

Defines recovery states and provider-specific retry rules. Enforces atomic cumulative run budgets and daily tenant budgets. Handles approval changes, revoked access, cancellation, and stale workers.

Staff (L6+/E6+)

Defines precise admission and external-effect guarantees. Explains where provider behavior limits those guarantees. Balances safe pauses against progress during failures and unknown charges. Extends budget and authority rules across concurrency, midnight, and regions.