Inside Assemble's Bet That AI Can Finally Fix Enterprise Software Deployment
A San Francisco startup claims its artificial intelligence agents completed in hours what the company says typically certifies Salesforce developers over months: earning the platform's All Star Ranger designation through 629 badges and 30 advanced challenges, with a 93 percent first-attempt pass rate.
Assemble, which emerged from Y Combinator this past summer, is automating one of corporate IT's most grinding chores. When businesses need to modify core systems like Oracle, SAP, or Salesforce, teams spend weeks tracing dependencies, writing integration code, validating results, and assembling documentation for audit committees. The company says its agents compress that timeline while generating what it calls an "agent trace" — a decision log that allows rollbacks if something breaks.
The technical demonstration matters less for the badges themselves than for what they suggest about automation reaching into skilled work. Salesforce Superbadges require multi-step problem-solving across real-world scenarios, not multiple-choice tests. That Assemble's system navigated them points to a capability that extends beyond rote tasks.
Consider a repair job the company published as a case study. An agent tackled a broken order sync between Salesforce and NetSuite by reading a ServiceNow incident ticket, tracing the failure through MuleSoft, identifying a missing field in NetSuite, quantifying how many deals were affected, fixing the sandbox logic, adding a regression test, and delivering eight validation checks across 25 tool calls. The output, according to Assemble, included 17 evidence artifacts packaged for a change advisory board.
The platform currently supports Oracle, SAP, Salesforce, Microsoft, ServiceNow, AWS, Workday, HubSpot, NetSuite, and MuleSoft. Changes get verified before deployment, with a rollback feature the company markets as "Time Machine."
How the System Works
Assemble's agents scope cross-system implementations, write configuration or code, validate results, and package change requests for human approval. The company combined two approaches: computer-use automation that controls interfaces like a human user, and direct calls to Salesforce's Metadata API for reliable deployment. That hybrid model attempts to balance UI flexibility with the auditability large enterprises demand.
The Trailhead certification run, detailed in a July blog post, started from a new account. The agent accumulated 371,300 points and completed 10 trails on its way to the All Star Ranger designation, which Salesforce uses to recognize developers who've mastered advanced skills. Execution time clocked in around 6.4 hours of recorded activity.
The company claims its agents work up to 40 times faster than manual processes and cut IT spending in half, with full pre-deployment verification. Assemble declined to name customers or provide third-party validation of those figures.
The Founders and the Timing
Aliyan Ishfaq, the CEO, graduated from Stanford's computer science program and previously worked at SAIL and LangChain. His co-founders bring similar pedigrees: Shaurnav Ghosh, the CTO, is a SAIL researcher with a Stanford master's degree who spent time at AWS and Apple. Shrish Janarthanan, chief product officer, studied computer science at Stanford and worked in cybersecurity at KPMG. The team numbered four as of an earlier snapshot, according to Y Combinator's directory, though that figure may have changed.
"We build agents for IT teams that manage enterprise systems like ERP, CRM, and HRIS platforms," Ishfaq wrote in the company's Y Combinator launch post.
The company arrives as enterprise software giants scramble to bolt AI onto their platforms. ServiceNow launched what it calls an "Autonomous Workforce" in May, deploying AI specialists for IT, CRM, and security workflows under an AI Control Tower. SAP unveiled its "Autonomous Enterprise" at its Sapphire conference the same month, showcasing Joule Agents alongside partnerships with Anthropic, AWS, Google Cloud, and Microsoft. Salesforce positioned Agentforce as an agentic enterprise platform at Dreamforce in September. Harness, which focuses on software delivery rather than business systems, announced autonomous worker agents in late June.
Whether Assemble's approach gains traction may depend less on technical prowess than on enterprises' willingness to let AI touch their most sensitive systems. The startup is hiring for sales development and enterprise strategy roles in San Francisco and remotely, a signal it's moving from demonstration to deployment.
The real test isn't badges on a Trailhead profile. It's whether IT directors will trust an agent to modify payroll integrations or customer data flows without a developer watching every step.
