The pitch sounds almost absurd: hand us your Salesforce migration, your SAP logic rewrites, your Workday configuration nightmares—the kind of enterprise IT work that typically consumes half a year and requires multiple meetings with Change Advisory Boards—and we'll finish it before your morning coffee gets cold.
Assemble, a newly minted startup from Y Combinator's summer cohort, is making exactly that promise. Three Stanford graduates think they've cracked a problem that has bedeviled enterprise IT departments since before cloud computing was even a phrase: the soul-crushing backlog of system modifications that require careful scoping, painstaking manual coding, and verification processes so exhaustive they make launching a spacecraft look breezy.
The company went live in mid-July, armed with autonomous agents designed to compress timelines from months to hours. But here's where it gets interesting—they're not just promising speed. They're promising the audit trails, rollback capabilities, and documentation rigor that actually make enterprise CIOs sign checks.
In the world of corporate IT, that's the real trick.
Automation for the Shadows
Most enterprise software automation targets the flashy stuff: customer service chatbots, marketing workflows, sales intelligence. Assemble is going after something less glamorous but far more consequential—the configuration tweaks, data migrations, and business logic updates that keep massive organizations running but rarely make headlines.
The product operates through what founder Aliyan Ishfaq calls a "control layer under every agent run." Translation: the agents don't just execute changes. They scope dependencies first, calculating what the company terms "blast radius"—industry speak for figuring out which systems and processes might explode if something goes wrong. Then they deploy changes from a unified workbench, handling the mechanics of code-writing, data-moving, and configuration-adjusting. Finally, they verify everything against shadow environments before touching anything in production.
A demo linked from the company's launch materials walks through migrating trade-show accounts from HubSpot to Salesforce, then setting up APEX automation for hot leads. Every step generates an audit trail. Every change can be reversed via snapshots.
This matters, perhaps more than the founders expected, in enterprise IT. Change Advisory Boards—those committees that control what touches production systems—demand documentation and rollback plans before approving modifications. Assemble positions its agents as CAB-ready by design, maintaining what it describes as a "reversible ledger" for business systems. If something breaks, teams can restore previous states.
Whether Change Advisory Boards will actually approve autonomous agent changes is another question entirely.
The Stanford Connection
Ishfaq, the CEO, studied computer science and AI at Stanford and worked at the Stanford AI Lab before a stint at LangChain. His writings from late 2025 focused on domain-specific coding agents, suggesting the product direction preceded the company's formal founding by several months.
Co-founder Shaurnav Ghosh also comes from Stanford's AI Lab, with previous stops at AWS and Apple. The third co-founder, Shrish Janarthanan, rounds out the Stanford trifecta—computer science degree, Mayfield Fellowship, and a cybersecurity internship at KPMG before joining Assemble.
The team remains at three people according to Y Combinator's directory, with primary partner Ankit Gupta overseeing their progress through the accelerator program.
That small headcount is either admirably lean or worryingly thin, depending on how you look at it. Building agents that understand Salesforce APEX, SAP ABAP, and Workday configurations at a level beyond simple API calls isn't trivial work. Neither is selling into enterprise IT departments, which move at geological speeds and ask questions that would make most startup founders uncomfortable.
The Fine Print
Enterprise buyers ask different questions than startups, and Assemble's legal documents acknowledge this reality head-on. The Terms of Service, updated in late May, state plainly that the product uses autonomous agents, AI output is experimental, and customers bear responsibility for verifying changes before production deployment.
There's also this: using Assemble requires granting API and credential access to third-party systems—Salesforce, NetSuite, Workday, the works. That's a significant trust proposition for IT departments that guard system access like nuclear launch codes.
The Privacy Policy, updated the same day, routes requests through what it calls "enterprise-grade AI model providers," specifically naming OpenAI and Anthropic. The company collects minimal data, it says, and operates under Delaware or California law—standard fare for venture-backed entities.
Security features emphasized on the website include blast-radius analysis, version control, rollbacks, and auditable traces for every agent action. These capabilities speak directly to the compliance and risk management concerns that shape IT purchasing decisions at organizations large enough to have Change Advisory Boards in the first place.
A Crowded Moment
Assemble is hardly alone in chasing the enterprise agent opportunity. Coralogix pulled in $200 million this summer for observability tooling to watch agents in production. NewCore raised $66 million around the same time to give agents digital identities within enterprise systems. OpenAI itself updated its Agents SDK in April with safety improvements aimed at enterprise deployments.
The infrastructure around autonomous software agents is maturing quickly. Whether the agents themselves are ready for prime time remains an open question.
Serval, focused on IT service management agents, raised $47 million last October. WitnessAI brought in $58 million in January to govern enterprise AI risk. Sierra—which targets broad enterprise automation—landed a staggering $950 million in May. The sheer volume of capital flowing into adjacent categories suggests investors see sustained demand for tools that make AI agents usable inside large organizations.
A report from SAP Insider earlier this year noted that ERP systems are becoming a starting point for AI projects. Enterprises want to layer intelligence onto existing infrastructure rather than replace it. A BCG paper from around the same time outlined practical design considerations for enterprise agents, emphasizing control mechanisms and clear boundaries.
Assemble's product addresses those themes directly. It doesn't ask IT teams to rip out Salesforce or SAP. It offers to do the implementation work inside those platforms, with guardrails attached.
What We Don't Know
Assemble hasn't disclosed pricing or packaging details. The website features a "Talk to Team" form rather than self-service pricing, signaling a demo-driven sales motion typical of early-stage enterprise software. Customer logos are absent. The company hasn't announced formal partnerships with the ERP, CRM, or HRIS vendors whose systems it supports.
No external funding round has been disclosed beyond the standard Y Combinator investment. The company launched days ago, making it too early for published case studies or third-party validation of agent performance in production environments.
That's a lot of unknowns for a product asking IT departments to trust autonomous agents with mission-critical systems.
The Central Question
Assemble is betting that enterprise IT leaders will delegate development work to autonomous agents if those agents come with the same rigor applied to human-driven changes: scoping, testing, documentation, rollback plans. The product treats enterprise systems as first-class integration targets rather than endpoints for simple API calls. The agents need to understand Salesforce APEX, SAP ABAP, and Workday configurations at levels that go well beyond basic read/write operations.
Whether that bet pays off depends partly on technical capabilities—accuracy, speed, handling of edge cases—and partly on organizational readiness.
Enterprise IT operates under constraints that don't apply to consumer products: compliance requirements, change approval processes, uptime commitments that border on religious observance. Assemble's architecture acknowledges those constraints. The harder question is whether enterprises are ready to delegate development decisions to software that carries an "AI output is experimental" disclaimer in its terms of service.
For now, the company is live, the demo is public, and three Stanford graduates are taking meetings with the CIOs and IT leaders who control enterprise software budgets. The agents are running. The audit trails are logging.
And somewhere, a Change Advisory Board is about to review its first AI-generated rollback plan. How that conversation goes will tell us more than any product demo ever could.
