The Pivot That Narrowed the Aperture
James Jiang scaled an AI coding platform to 1.5 million users, according to James Jiang's LinkedIn profile, before co-founding MSPilot, the Y Combinator‑backed startup building an AI infrastructure layer for IT service providers. That traction, and the decision not to raise after a first Y Combinator run, shapes how Jiang and Paul Li are building MSPilot, which entered Y Combinator's Fall 2025 batch with three people and a thesis: managed service providers need governed AI agents, not another chatbot.
The through-line is Li's observation that AI helps users start fast but leaves them stuck finishing. He saw it first at EasyCode, the VS Code and JetBrains extension he and Jiang built after dropping out of the University of Waterloo. He saw it again at OpenBuilder, where the team observed that over 80 percent of non-technical vibe coders couldn't complete their projects. Now he's betting the same dynamic plays out inside the fragmented IT-services market.
Li and Jiang's first venture together was Mirage VR, a virtual-reality gaming startup. When that stalled, they pivoted in 2023 to EasyCode, which grew to 1.5 million installs without a sales team. Y Combinator accepted them, but Li chose not to fundraise; the company was growing and had already closed a round. He later called that a mistake. Competition from Cursor, Copilot, and Replit intensified, and EasyCode's momentum plateaued. "Li said he regretted that decision as competition within the field intensified," the MoneyTimes reported in April 2026.
Jiang brings an Amazon software-engineering background and a self-described focus on "unlocking human potential with AI." Li describes himself as a three-time founder with a computer-science degree from Waterloo. Their LinkedIn profiles note they "previously built AI coding tools used by over 1.5 million vibe coders." The phrasing matters: they frame EasyCode's users as "vibe coders" (non-technical builders who prompt rather than program) because that cohort became the target for OpenBuilder.
"They kept seeing the same pattern: AI helps you start fast, but people get stuck finishing."
That pattern drove OpenBuilder's design. Instead of credit-based pricing that penalizes iteration, Li adopted a fixed subscription for unlimited generation, with an escape hatch: pay for a human developer who already has your project context when the debugger fails. The platform runs on open-source models from Z.ai and DeepSeek, which Li argues are cheaper and more flexible than closed alternatives. "As AI costs fall over time, Li says pay-per-use business models will become untenable," the MoneyTimes noted. His bet: "Our bet is that over time, users are going to become educated and aware enough of the landscapes that eventually it will shift to more fixed pricing."
OpenBuilder positioned itself as the open-source alternative to Lovable and Replit: no lock-in, built on VS Code, export your codebase anytime. The team plans to open-source the codebase itself. But the deeper lesson Li carried into MSPilot isn't pricing. It's distribution. EasyCode reached developers directly. OpenBuilder targets hobbyists and aspiring entrepreneurs. MSPilot targets managed service providers, the firms that already manage IT for small businesses and control the procurement relationship. Li calls it a "genuine blue-ocean space" where no incumbent has built the infrastructure layer, yet small businesses already pay thousands per month for AI delivered this way.
The founder pedigree is not a trophy. It's a set of scar tissue: a VR pivot, a missed fundraising window, a pricing-model pivot, a distribution-channel pivot. Each iteration narrowed the aperture from "developers" to "non-technical builders" to "the channel that serves small businesses." MSPilot is the narrowest aperture yet — and the first where the go-to-market motion doesn't require convincing end users to adopt a new tool. The MSPs already have the contracts. They just need the rails.
YC Backing and the Architecture Bet
MSPilot entered Y Combinator's F25 batch as a three-person team based in San Francisco. The accelerator's stamp matters because the founders are not first-time builders; Jiang previously achieved that scale, a track record that signals an ability to ship product at velocity. YC's decision to back the company at this stage reflects a bet on that execution history as much as on the market thesis.
The mission is specific: build the AI infrastructure layer for the IT services industry. That means giving managed service providers (MSPs) a platform to deliver governed AI to the small businesses they already manage, and to build and run AI agents for those clients on the same rails.
"We're a Y Combinator-backed startup building that infrastructure layer: the platform IT providers use to deliver governed AI to their clients and to deploy AI agents for them on the same rails."
The architecture splits into two tightly connected halves. The first is AI access and governance infrastructure, a gateway layer. It runs a multi-tenant AI gateway that presents an Anthropic/OpenAI-compatible API layer. Popular AI tools such as Claude Code, Codex, and Gemini CLI are pinned to it via managed endpoint configuration. Every request binds to a real employee identity through Microsoft Entra ID using OIDC/SAML federation, with per-tenant isolation across hundreds of end-client organizations. The gateway handles usage metering, pooled credit management, per-tenant billing, audit logging, policy enforcement, and sensitive-data controls at the request layer. An endpoint agent deploys and pins tool configurations on managed devices and detects drift.
The second half is the agent runtime and builder infrastructure. Agents execute in a tenant-isolated runtime with scoped, vaulted credentials that inherit the permissions of the employee or role they act for. A managed MCP (Model Context Protocol) server registry provides sanctioned connectors to business systems, scoped per tenant. Observability comes through replayable execution logs, approval workflows for actions, and blast-radius controls. On top sits a builder layer, including templates and an authoring flow that let non-experts assemble agents that are safe by construction.
This is infrastructure work in the traditional sense: multi-tenancy, identity, credential security, gateways, billing, with an AI-native product layered on top. The platform lets an MSP author an agent once in a single tenant, then roll it out across the whole book with per-tenant configuration. Usage is metered per tenant and per agent; the MSP sets the markup and the line item lands cleanly on the invoice they already send. Margin stays with the MSP, not the model vendor.
Horizontal templates cover voice agents for 24/7 front-desk coverage, internal QA bots that audit tickets and calls against policy, and document processing that turns paperwork into structured data. Vertical templates target legal intake and matter summaries, healthcare patient scheduling and compliant records handling, manufacturing order processing and shop-floor knowledge lookup, accountants' reconciliation and tax-season triage, non-profit donor outreach and grant drafting, and education enrollment support. Integrations plug into both the MSP's stack and their clients' stacks.
Security and maintenance are fully managed: per-tenant roles, approvals, and audit logs; tenant-isolated data with encryption in transit and at rest; SOC 2 and HIPAA readiness. The platform runs the VMs, hosting, tokens, and patching, backed by automation engineers who build and maintain agents alongside the MSP.
The Founding Engineer Who Must Ship the Rails
MSPilot's three-person team is moving fast. The Y Combinator F25 company, founded in 2026 by Jiang and Li, lists two open roles (one in sales, one in engineering), but the engineering posting reads like a founding-engineer manifesto. The role, titled Founding Engineer - AI Infra (Hybrid, Lviv), carries a $60K–$100K salary band and 1–2% equity, with U.S. citizenship or visa explicitly not required. The posting asks for three-plus years of experience, though the detailed requirements cite five-plus years of backend or platform engineering. That tension (senior scope, early-stage title) mirrors the company's bet: the first engineering hire will own "large parts of the platform architecture" from day one, working directly with the founder, no legacy code, no management layers.
The LinkedIn headline for co-founder James Jiang makes the bar explicit: "Hiring 10x Engineers · Building the Agentic Automation Platform for MSPs." The job description leans into the same language, "founding-level equity - meaningful ownership, not a token grant," and promises a hiring process under two weeks, with equity vesting backdated to the start date. Compensation is benchmarked to senior/staff levels and discussed on the first call. Hybrid work anchors in Lviv with core overlap hours for the San Francisco-based founder, plus a hardware/co-working budget and paid time off.
What that engineer will build is the gateway, runtime, identity, MCP registry, observability, and builder layers detailed in Section 2. The hiring push is the execution lever. The sales role rounds out the pair, but the founding engineer is the keystone: the person who turns the spec into a shippable product the early MSP pilots can deploy. The company's blue-ocean claim — no incumbent has built this yet — only holds if the engineering hire ships the gateway and runtime fast enough to lock in the demand signal.
Demand Signals: From One Percent to Twenty in Half a Year
Demand for AI agents from managed-service providers has accelerated faster than most vendors predicted. Six months ago, roughly 1 percent of MSPilot's target clients were asking for agents. Today that figure sits at 7 percent. The company projects it will reach 20 percent within a year, a sevenfold increase in half a year. The data comes from MSPilot's own tracking of inbound requests across its early customer conversations.
Law firms have emerged as the fastest-adopting vertical. Their work is high-value, document-heavy, and automates cleanly. Dental practices, accounting firms, and field-service operators follow close behind. These four segments form the wedge for MSPilot's first template library. The pattern is consistent: professional-services businesses that bill for expertise and drown in repetitive document processing see immediate ROI from an agent that handles intake triage, contract review, or appointment scheduling.
The revenue model reflects that urgency. MSPilot estimates $2,000 to $5,000 in new monthly recurring revenue per client for the IT provider. A typical provider manages 25 to 250 SMB tenants. Stack the revenue lines (subscription, token markup at 3x, infrastructure markup at 3x) and a single provider booking 184 tenants could see projected MRR of $4,920 per month, or roughly $59,000 annual run-rate. Per-tenant revenue lands around $19,680 per year.
| Revenue Component | Monthly per Tenant | Markup | Monthly Contribution (3 tenants) |
|---|---|---|---|
| Subscription | $1,500 | — | $4,500 |
| Token markup | $90 | 3x | $270 |
| Infra markup | $50 | 3x | $150 |
| Total MRR | $4,920 |
Assumes roughly $45 in LLM tokens and $25 in infrastructure per tenant per month. Actual numbers vary with usage.
The platform's multi-tenant architecture makes those numbers possible. One agent (say, an intake-triage workflow) gets authored in a single tenant and then rolls out across the entire book with per-tenant configuration. MSPilot's demo environment shows that same agent deployed across six of 184 tenants: Northwind Law, Cedar Dental, Atlas Books, Vertex HVAC, Clearwater Clinic, and Summit Agency. Each tenant sees its own roles, approvals, and audit logs. Data stays isolated, encrypted end-to-end, with SOC 2 and HIPAA readiness baked in.
Governance is the unlock. Providers cannot resell generic AI wrappers to regulated clients. They need per-tenant role-based access, human-in-the-loop approvals, and a full audit trail that survives scrutiny. MSPilot bakes those controls into the control plane so the provider does not have to build them from scratch.
The integrations list reads like an SMB's entire stack: QuickBooks, Salesforce, HubSpot, Slack, Google Workspace, Microsoft 365, Shopify, Stripe, Zendesk, Notion, Asana, Mailchimp, plus any API. That breadth matters. An agent that cannot reach the client's CRM or accounting system is a demo, not a product.
Pax8's MIP Playbook captures the economic shift in one line: "When an 8-person consulting firm can deliver what once required 50, the economics of professional services shift overnight." MSPilot's pitch is that the IT provider owns the relationship and the access — the trust, the keys to every tenant's systems — and now, with this infrastructure layer, the ability to turn that position into a recurring revenue line. The provider builds once, meters per tenant, marks up the cost, and pushes clean line items to the PSA and billing tools already in use. A June invoice for Northwind Law's intake-triage agent shows 12,400 runs, $420 in metered cost, an 180 percent markup, and $1,180 billed, yielding a 64 percent net margin.
The Managed Intelligence Provider (MIP) model only materializes when an MSP can deploy a governed agent to Client A on Monday, clone the policy to Client B on Tuesday, and bill both on Wednesday without a professional-services engagement. That operational rhythm is the real product. Everything else is scaffolding.
Clients are asking. They cannot always describe what they need. They don't vibe-code. The use case has to be surfaced for them. MSPilot's platform does that surfacing, then handles the deployment, governance, and billing. The demand signal is no longer theoretical. It is showing up in inbound volume, vertical concentration, and the revenue math that makes a provider's next hire an AI engineer instead of another help-desk tech.
The Market's Fault Lines and What Must Hold
The managed services market MSPilot targets is not a monolith; it splits into distinct tiers with different buying motions. CRN's MSP 500 list separates providers into the Elite 150 serving midmarket and enterprise accounts, the Pioneer 250 focused on small businesses, and the Security 100 specializing in cloud-based security. MSPilot's AI control plane for the MSP channel must work across this stratification. A platform built for a 50-person Pioneer MSP managing 200 endpoints behaves differently than one serving an Elite 150 firm running hybrid infrastructure for 5,000 seats. The company's early traction with small-business IT firms suggests it is starting at the Pioneer layer, but the revenue upside sits upstream.
Implementation friction remains the loudest signal in the market. CompareMSP's analysis of negative MSP reviews shows 69 percent cite delayed ticket handling and slow response times, symptoms of operational debt that AI promises to relieve but often compounds during rollout. Worksent's August 2025 survey of MSP tooling found providers struggle with implementation, integration, and the cost of scaling even as they adopt Atera Autopilot, ConnectWise Automate, SuperOps.ai, and Microsoft Copilot. Traditional rule-based monitoring, built on static thresholds, fails in the hybrid cloud-first environments where most SMB clients now operate. MSPilot's architecture — two tightly connected halves for governance and agent execution — bets that a unified control plane reduces that integration tax. The bet is unproven at scale.
Competition is not absent; it is adjacent and deep-pocketed. Microsoft's 2026 roadmap includes Azure AI Agent Service, Dynamics 365 AI features, and Copilot Studio, all targeting the same "build and run AI agents" workflow MSPilot pitches. Darktrace, Splunk Security Cloud, and LogRhythm already own the AI-driven security layer for many MSPs. ServiceNow ITSM and Atera embed automation inside existing PSA/RMM stacks. MSPilot's founders argue a genuine blue-ocean space exists because no incumbent has built the governance layer that lets an MSP monetize agents across every SMB client on shared rails. That claim holds only if the governance problem — policy enforcement, audit trails, multi-tenant isolation — is harder than the agent-building problem. Microsoft's own "Governing AI agents at scale" guide, published to adoption.microsoft.com, suggests the hyperscaler treats governance as a first-class product surface, not an afterthought.
Talent acquisition compounds the technical risk. CRN's 2025 MSP 500 executive survey lists difficulty in hiring skilled employees alongside margin pressure and cybersecurity threats as top business challenges. MSPilot itself is hiring a founding engineer to own significant portions of the platform architecture while the founding team of three splits product, sales, and infrastructure. The first engineering hire will inherit a codebase that must multi-tenant from day one, enforce governance policies across heterogeneous client environments, and integrate with the RMM/PSA tools each MSP already runs.
The market's next phase, per CustomGPT's November 2025 analysis, centers on three trends: Adaptive AI Systems that retrain on live telemetry, Multimodal Intelligence that fuses logs, tickets, and network flows, and AI Integration Across the Stack that eliminates point-solution sprawl. MSPilot's roadmap aligns with all three, provided it can ship the governance primitives first.
The Rails Are the Product
As noted earlier, the MSPs hold the contracts and just need the rails. That line, delivered in the first section, described a distribution insight. Five sections later, it describes a technical reality: a gateway that binds every request to a real employee identity, a runtime that scopes credentials to the role, a builder that lets non-experts assemble agents safe by construction. The founding engineer who takes the Lviv role will not be building a demo. They will be laying the track that lets an MSP push a governed agent from Northwind Law to Cedar Dental to Atlas Books — Monday, Tuesday, Wednesday — without such an engagement each time. The blue ocean is not the market. It is the control plane that makes the market operable.
Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.