Assemble agent · Use case · Development

Ship partner-facing Apex web services

Aero Technica Solutions is launching a drone product line and needs three partner-facing APIs: a permissible-fly-zone lookup, a lost-device kill switch, and a pilot certification check. The Assemble agent shipped all three as Apex web services in 7 min 16 sec against the ~4 hr 40 min Trailhead estimate. Every acceptance check passed against the live org, 3 of 3.

0:00 Running
Send a follow-up…
01

What this does

Three services, one org. ProductZoningService answers GET /ProductZoning/{productCode} with the permissible fly zone for the country named in the CountryCode header. AssetService exposes POST /lost — a kill switch that retires a lost drone and files a claim when comprehensive coverage is active. CredentialVerificationService confirms a pilot's certification status over SOAP.

Before touching anything, the Assemble agent read the org: instance assemble60-dev-ed.develop.my.salesforce.com on API v67.0 (Summer '26), then full describes of Product_Geo_Mapping__mdt, Asset, Insurance__c, Claim__c, and Contact_Certification__c. Fly-zone rules, device serials, coverage, and pilot credentials were already modeled.

That inspection settled the plan: no schema changes, just services over data that already exists. The whole change is six ApexClass records, three services and three test classes, each reversible with a single recorded DELETE.


02

How it works

Salesforce TATooling API ARApex REST
  1. 01Read the brief. The business requirement names the three services, their endpoints, and the acceptance checks that will judge the final org state.
  2. 02Inspect the org. Describe calls against the newest API version map what already exists. No code is written until the data model is understood.
  3. 03Stage and approve. All six classes are compiled, validated, and staged as one reversible unit. The staged change is approved before deployment.
  4. 04Deploy via Tooling API. Six POST /tooling/sobjects/ApexClass calls, each answered 201 Created, each with a recorded revert path.
  5. 05Test until passing. The first suite run fails on a real org rule; the Assemble agent rebuilds the fixture and re-runs to a full pass above 85% coverage.
  6. 06Verify. The acceptance checks inspect the org itself. Three checks, three passes.

Same shape a senior Salesforce engineer would use: read the org before writing code, stage one reviewable change, let the failing test explain the org's rules, and don't call it shipped until every acceptance check passes.


03

Outcomes you can verify

The results aren't self-reported. The acceptance checks inspect the org itself. All three pass by reading the deployed classes and the records they touch. Claims alone don't count; org state does.

The endpoints answer for themselves: GET /services/apexrest/ProductZoning/GC1020 with CountryCode: US returns "Recommended", and POST /lost with an unknown serial returns "No device found." Every persistent change sits in the ledger with a revert path — six creates, six recorded deletes, nothing else touched. The full run: 30 steps, 0.95M input / 27K output tokens.

Duration
7:16
wall clock, brief to verification
Trailhead estimate
~4 hr 40 min
for the same work by hand
Speed-up
38.5×
same work, same org
Steps
30
reads, writes, tests, checks
Checks
3/3
PASS, run against the live org
Org changes
6
ApexClass creates, all reversible
org-inspection.md deployed-classes.csv, 6 record ids test-results.csv, 3 suites + coverage acceptance-checks.csv, 3 of 3 PASS revert paths, 6 × DELETE