The CRM migration was supposed to take three months. Nine months in, the consultants are still untangling field mappings between Salesforce and NetSuite, burning through budget while business users improvise workarounds in Excel.
It's a story so common in enterprise IT that it barely registers as news anymore. What does register: a three-person team from Stanford claims they've built AI agents that can execute these projects autonomously, complete with audit trails that might actually survive a compliance review.
Assemble's YC launch post was published in mid-July 2026, emerging from Y Combinator's Summer 2026 batch with a proposition that sounds either obvious or impossible, depending on who you ask. Their software deploys agents directly into enterprise systems—Salesforce, SAP, Workday, ServiceNow—and turns them loose on the integration and customization work that typically gets outsourced to consultancies or buried in IT backlogs. Point the agent at a migration project, the founders say, and it will scope dependencies, modify business logic, validate changes, and produce rollback snapshots if something goes sideways.
The demo they published alongside the YC launch shows an agent migrating trade-show accounts from HubSpot to Salesforce, then configuring APEX automation to flag hot leads. The system executes the migration, builds the workflow, and logs every action. When the founders deliberately revert a change mid-demo, it rolls back to an earlier snapshot. Version control for enterprise software configuration, not just code.
Whether that demo translates to production environments at scale is another question entirely.
The Grunt Work Problem
Assemble markets itself as "agents for Enterprise IT," which is perhaps a more palatable pitch than "software that replaces your SAP consultants." The product targets what one might charitably call the unglamorous layer of enterprise software: syncing data between systems, deduplicating vendor records in NetSuite, propagating organizational changes from Workday to NetSuite after a restructuring, fixing broken field mappings after an ERP upgrade.
The workflow follows a three-step pattern the company calls scope-run-verify. First, the agent analyzes dependencies across target systems—what breaks if this field mapping changes? Then it executes modifications. Finally, it validates outcomes. Every action gets logged in what Assemble describes as an "agent action ledger," which IT teams can inspect or replay later. If a change triggers downstream failures, administrators can roll back to a previous system state using built-in snapshots.
Assemble's website displays a familiar constellation of enterprise logos: Salesforce, NetSuite, MuleSoft, ServiceNow, Oracle, SAP, Microsoft, AWS, Workday, HubSpot. The exact split between generally available connectors and roadmap ambitions remains unclear. Published use cases include migrating Google Sheets and Notion data into HubSpot, syncing Salesforce records to NetSuite, and cleaning up duplicate vendor entries in financial systems—the kind of work that keeps mid-tier consulting practices in business.
A July 24 blog post from the team detailed their infrastructure for Salesforce org version control, explaining how they combine UI automation with Salesforce's Metadata API to achieve what they claim are deterministic rollbacks. The post mentioned internal benchmarks from Salesforce Trailhead scenarios. Customer case studies or pilot metrics? Not yet published.
Stanford AI Lab to YC
All three founders came through Stanford's computer science program and overlapped at the Stanford AI Lab, which has become something of a startup incubator in its own right. CEO Aliyan Ishfaq previously worked at LangChain and SAIL, focusing on applied AI. His personal site mentions an earlier product, ParaAI, that reportedly reached over 100,000 users. Co-founder Shaurnav Ghosh also worked as a SAIL researcher, with stints at AWS and Apple, and has published work on using large language models for brain MRI analysis. The third co-founder, Shrish Janarthanan, spent time at KPMG working on AI and cybersecurity projects and was selected for Stanford's 2025 Mayfield Fellows cohort.
The company is legally incorporated as Standard Interaction, Inc.—a detail visible only in the website footer and mirrored domain. According to the YC directory, the team size remains at three as of early August. No open job postings. Pricing information isn't public; prospects get directed to a "talk to team" form for demos, suggesting the company is still calibrating its sales motion.
A Crowded, Suddenly Urgent Market
Assemble's timing, at least, is defensible. The launch lands in what Gartner, in an August 2025 forecast, predicted would be a breakout year for enterprise AI agents in 2026. The research firm estimated that 40% of enterprise applications would integrate task-specific agents by the end of 2026, up from less than 5% in 2025. An S&P Global report from November 2025 found that 58% of organizations were already pursuing agent capabilities, though many struggled with infrastructure requirements beyond chat-based systems.
The incumbents didn't wait around. Salesforce made its "Agentforce" setup tools generally available in June, according to the company's admin blog. ServiceNow updated its "Now Assist" platform in June and July with agentic workflows for IT service management. SAP announced its AI Agent Hub and expanded Joule agents in May, then partnered with Google Cloud in April on multi-agent orchestration. Google itself launched Gemini Enterprise in late April as a unified platform for agent development with third-party system connectors. UiPath and Automation Anywhere both unveiled agentic automation platforms earlier in the year, touting cost reductions and service-desk automation.
That wave of platform releases creates both competition and, possibly, opportunity. Enterprises are already being sold agent capabilities by their ERP and CRM providers, but those tools focus on end-user productivity—smart assistants for sales reps or service agents answering tickets. Assemble is betting that IT teams want agents for different work: the backend plumbing, the integrations and migrations and configuration updates that accumulate in Jira backlogs and rarely get prioritized until a system breaks.
A July research paper on arXiv proposed a similar concept—"Agentic ERP" using role-aligned LLM agents with human oversight for ERP operations—suggesting the idea has at least some academic traction beyond the startup's pitch deck.
The Execution Problem
Assemble's core promise is auditable automation. The company emphasizes controls: rollback snapshots, blast-radius analysis before changes, action-by-action logs. That positioning makes sense for an audience conditioned to treat production ERP and CRM systems like live munitions.
But as of August 2026, there are no publicly available customer case studies or third-party validations of those safety claims.
The technical challenge here is non-trivial, and perhaps that's understating it. Enterprise systems like SAP and Salesforce are notoriously complex, with thousands of configuration points, custom code, and integrations that break if you look at them wrong. Agents that can reliably scope dependencies across multiple systems, execute changes without collateral damage, and generate audit trails that satisfy governance teams would represent a significant leap beyond current AI tooling. The company's Salesforce blog post offers a glimpse of the infrastructure required—custom version-control systems, deterministic rollback mechanisms, integration of metadata APIs with UI automation. Those claims await real-world stress testing.
The market also remains, to put it gently, undefined. IT leaders comfortable letting agents modify production Salesforce orgs or SAP instances are scarce. Especially when the vendor is a three-person startup with no reference customers. Assemble will likely need to prove itself in sandbox environments and low-risk use cases before enterprises trust it with mission-critical systems. The absence of pricing information suggests the company is still figuring out product-market fit.
Still, the underlying premise is hard to dismiss. If the months-long ERP customization backlog is a universal problem—and ask any mid-sized company with a pending NetSuite migration—then software that can compress timelines while maintaining audit compliance has a wedge. Whether Assemble's agents can actually deliver on that promise at scale? That's the question early pilots will answer, assuming the founders can convince anyone to run them.
