Pinecone is Sierra's shared agent interface for employee work. It routes requests to a model, harness, and environment, runs sessions on Agency, and connects to company systems through an MCP Gateway.
Employee prompt to a streamed cloud session with brokered pull requests.
Start a session from a prompt
A prompt creates an isolated pod with a checked-out repository, warm build cache, sidecars, and a harness; the session streams its work back for watching, interrupting, or redirecting[Source 1]
Broker its own pull requests
A session opens its own pull requests and watches its own checks[Source 1]
Where people stay involved
People review, approve, or guide the agent’s work here.
Reported: Sessions stream to the requester, who can watch, interrupt, or redirect; PR babysitting escalates to a person only when judgment is required.
MCP Gateway connected to 45 systems; a network proxy decides whether privileged requests proceed and injects credentials after approval[Source 2]
Knowledge
Durable sessions over disposable environments; Redis Streams[Source 1, Source 3]
Context management
Durable session state separated from ephemeral compute[Source 3]
Credentials
Gateway enforces the invoking employee's access at tool-call time and isolates customer data; the harness never holds the real secret[Source 2, Source 3]
One gateway spanning 45 systems, enforcing employee permissions at call time[Source 2]
Agency sandbox layer
Recoverable K8s runners with a network proxy for credential injection[Source 3]
Intent-based routing
Routes model and environment by request intent, independent of the tool layer[Source 1]
Validation and failure handling
Babysit the pull request
A monitor sub-agent answers machine-review comments, fixes failing tests, rebases on conflicts, and pings a person in Slack when a decision needs a human[Source 1]
Reported observations
These results come from the sources. Notes below each result explain what it covers and any reported limits.
Headline claim
More than 75,000 sessions created by 600 people in the month preceding the report[Source 1]
Reported by
Sierra
Scope
Pinecone users and sessions in the month preceding the report; calendar month unspecified
The source does not report the denominator of this figure.
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Pinecone sessions created by Sierra employees in the month preceding the report
Key observation
70% of company PRs opened through Pinecone in the month preceding the report[Source 1]
Reported by
Sierra
Scope
Company PRs opened through Pinecone in the month preceding the report
Denominator
Sierra PRs opened in that month; not merged PRs
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Sierra pull requests opened through Pinecone in the month preceding the report
Key observation
96% of engineering uses Pinecone; the majority of the company works in it every day[Source 1]
Reported by
Sierra
Scope
Engineering use of Pinecone and company-wide daily use at report time
The source does not report the denominator of this figure.
Observed in 2026
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Engineering adoption of Pinecone and company-wide daily use
Key observation
Usage has tripled every month since April while costs fell[Source 1]
Reported by
Sierra
Scope
Pinecone usage growth since April and reported falling costs; no absolute figures published
The source does not report the denominator of this figure.
Observed in 2026
About this result
Category
Adoption output
Evidence type
Reported measurement
What it covers
Pinecone usage growth since April with reported falling costs
Lessons
Pinecone presents one entry point while a classifier chooses the repository, environment, harness, and model for the task.[Source 1]
Sierra wraps Codex and Claude Code behind a common runner protocol and can change routing defaults centrally.[Source 1]
Sierra's gateway audits customer-data access and blocks sensitive access spanning multiple customers in a session; some CLI access uses separate read-only tokens.[Source 2]
Catalog interpretation: Sierra's session total measures use of Pinecone; it does not by itself measure time saved or the quality of completed work.[Source 1]
Reported: The three pre-existing primitives are reusable mechanisms; the run is prompt to session to brokered pull request.
human involvement
Reported: Sessions stream to the requester, who can watch, interrupt, or redirect; PR babysitting escalates to a person only when judgment is required.
implementation
Reported
validation
Reported: The documented check is the babysitting loop: watched CI checks, fixed failing tests, and answered machine-review comments.
The retrospective describes shared adapters and configurable defaults; it supports interchangeable harnesses without claiming every model fits every task.
The source reports session volume and centralized cost oversight, but provides no method for converting each session into time saved or a quality measure.
