Vishal Sikka’s Hang Ten is taking aim at the old IT services model with AI agents

Vishal Sikka knows the IT services business from the inside. He ran Infosys, spent years building enterprise software at SAP, sat on Oracle’s board, and later founded VianAI. His new company, Hang Ten Systems, is aimed at the labor-heavy model that made Infosys, TCS, Wipro, Accenture, and Cognizant central to large enterprise technology work.

Hang Ten has raised a $32 million seed round led by Mayfield, with strategic backing from Aramco Ventures and participation from angel investors. Yahoo co-founder Jerry Yang is on the board. The company says it helps enterprises continuously build, change, and operate software using AI-driven development and automation.

That’s broad, but the market is broad. Large companies spend heavily on software customization, integrations, migrations, maintenance, testing, support, reporting workflows, and internal tools. Much of that work still runs through consultants and offshore delivery teams. Hang Ten is betting that AI agents, reusable software “skills,” and domain-specific engineering teams can reduce the amount of human labor needed to ship that work.

The notable part isn’t another startup promising AI-generated code. That category is crowded. The notable part is Sikka’s target: the economics of enterprise services.

The pitch: AI-native delivery for enterprise software

Hang Ten describes itself as an enterprise AI services company built around agentic code generation, reusable AI skills, and domain expertise. Stripped down, the model appears to combine four pieces:

• AI agents that generate, modify, test, and document code across enterprise systems.
• Reusable “skills” for repeatable work such as SAP integrations, data pipeline generation, reporting workflows, compliance checks, test generation, and deployment tasks.
• Forward-deployed engineers who work close to customers, translate messy business requirements into technical work, and supervise delivery.
• Domain experts who keep the system from producing naive implementations in regulated, legacy-heavy environments.

Hang Ten says it’s already working with Siemens Gamesa Renewable Energy and Fresenius on AI-native project delivery. Mayfield managing partner Navin Chaddha told TechCrunch the company “just got started a month back” and already has customers. For a seed-stage startup selling to enterprises, that’s a meaningful signal. Enterprise procurement rarely moves quickly, especially when a vendor needs access to codebases, internal systems, data flows, and operational processes.

The company is based in the Bay Area and is hiring across delivery, engineering, sales, and leadership, with plans for global expansion. Its early team includes people who have worked with Sikka across SAP, Infosys, and VianAI, including CTO Navin Budhiraja, chief design officer Sanjay Rajagopalan, and Tao Liu, senior vice president of forward deployed engineering.

That last title matters. “Forward deployed engineering” is the Palantir-style model where technical staff work directly with customers to turn vague operational needs into working systems. For AI services, that role becomes even more important. Enterprise AI projects usually don’t fail because nobody can call an LLM API. They fail because the requirements are fuzzy, the data is ugly, permissions are brittle, and nobody agrees what “done” means.

Why IT services are exposed

Traditional IT services scale through people. Win a large contract, assign a team, add managers, bill hours or milestones, repeat. Automation has always been part of the business, but the commercial engine is still tied to utilization rates and delivery capacity.

AI puts pressure on that model because a lot of software delivery is repetitive and semi-structured:

• translating specs into boilerplate code
• generating API clients and integration glue
• writing unit and regression tests
• converting legacy logic into newer frameworks
• creating dashboards and reporting layers
• scanning code for common security issues
• documenting old systems
• triaging support tickets
• handling CRUD-heavy internal applications

AI won’t safely replace senior engineers on complex systems. It can, however, compress a large amount of billable work.

Jefferies analysts argued earlier this year that IT services could be among the first sectors to face meaningful AI disruption. That concern has been hanging over the industry. Infosys shares are down more than 35% this year, and investors are asking whether generative AI will reduce demand for outsourced engineering labor or expand it by making more software projects economically viable.

Infosys chairman Nandan Nilekani has argued for the expansion case. Infosys has told investors that “AI-first services” could represent a $300 billion to $400 billion market by 2030. In that version, enterprises need outside help modernizing systems for AI, creating a new wave of demand.

Both outcomes can exist at the same time. AI may increase the number of software projects while reducing the labor required for each one. That’s awkward for incumbents. If pricing stays tied to hours, margins get squeezed. If pricing shifts toward outcomes, services firms have to prove they can deliver faster without damaging their own staffing model.

Hang Ten is built around that second assumption from day one.

The technical problem is messier than the pitch

Agentic software delivery sounds clean in a funding announcement. Inside a large company, it gets ugly fast.

Most enterprises don’t have one neat stack. They have SAP, Oracle, Salesforce, ServiceNow, Microsoft systems, custom Java and .NET applications, Python data pipelines, mainframe interfaces, dozens of data stores, and years of undocumented business logic. Access control is fragmented. Environments differ. Test coverage is uneven or missing. CI/CD may exist for newer systems while older systems still depend on manual release rituals.

AI agents can help, but only with serious guardrails.

Technical leaders will care about the unglamorous details:

• Can the system inspect and modify large codebases without losing context?
• How does it handle private dependencies, internal APIs, and undocumented workflows?
• Does it generate tests that catch real business regressions, or only shallow cases?
• Can it work inside existing CI/CD and change-management processes?
• How are secrets, credentials, customer data, and proprietary code protected?
• Who approves production changes?
• What happens when an agent produces plausible but wrong logic?

That last point remains the tax on every serious AI coding product. LLMs are far better at code than they were three years ago, but enterprise delivery needs correctness, traceability, and accountability. A generated function that passes a narrow test suite can still violate a business rule buried in a decade-old workflow.

That’s why Hang Ten’s human layer may matter as much as its automation layer. The company can talk about AI-native services, but enterprise customers will still want named people accountable for outcomes. Someone has to sign off on architecture, security, compliance, and production readiness.

Where Hang Ten has room to stand out

Sikka’s background gives Hang Ten credibility in rooms where most AI coding startups would struggle. CIOs and enterprise architects know the pain of long implementation cycles. They also know the cost of failed modernization programs. A founder who has run Infosys and built enterprise software at SAP can speak to those problems without sounding like a tool vendor guessing at enterprise reality.

Hang Ten also appears to be selling managed AI-assisted delivery to large customers rather than a self-serve developer tool. That fits enterprises that want completed work, not another IDE plugin.

AI coding assistants such as GitHub Copilot, Cursor, Windsurf, and enterprise agent frameworks have already changed how individual developers work. But tool adoption inside a company doesn’t automatically change delivery economics. Developers may write code faster while the organization still spends the same amount of time on requirements, review, testing, security approvals, integration, and release management.

A services model can address the full workflow. Hang Ten can package AI agents, engineering practice, project delivery, and domain knowledge together. If it works, customers won’t care much whether the output came from an agent, a human engineer, or both. They’ll care whether the migration finished in six weeks instead of six months, whether defects dropped, and whether the invoice made sense.

That’s the bet investors are backing. Mayfield’s argument is that traditional services firms scale linearly with headcount, while Hang Ten’s model improves with each project as reusable skills accumulate. In theory, every customer engagement creates assets that make the next engagement faster.

In practice, reuse is hard. Enterprise services firms have promised reusable components for decades. The promise often breaks because every customer insists its processes are unique, and sometimes they are. AI may make templates easier to adapt, but it doesn’t remove organizational complexity.

VianAI’s shadow

Hang Ten is Sikka’s second major AI startup after leaving Infosys. His previous company, VianAI, emerged from stealth in 2019 with $50 million in seed funding and later raised $140 million in a 2021 round led by SoftBank Vision Fund 2. VianAI focused on enterprise AI applications and analytics tools for business decision-making.

Mayfield says Hang Ten is aimed at a different market. That distinction matters. Enterprise analytics and decision-support tools have been crowded for years, with buyers often stuck between consulting-heavy deployments and dashboards that don’t change operational behavior. AI-assisted software delivery is a clearer pain point. Companies already spend money on it, outsource much of it, and measure timelines, costs, and defects.

Still, VianAI is relevant context. Big enterprise AI visions are difficult to turn into repeatable products. Sales cycles are long, customization creeps in, and “platform” stories can drift into consulting projects with software attached. Hang Ten will need to avoid becoming a traditional services firm with AI branding.

The hiring plan suggests the company accepts that services will be central. That’s not a flaw by itself. Many durable enterprise companies start with heavy customer involvement. The question is whether Hang Ten can turn delivery work into reusable automation assets rather than staffing projects with better internal tools.

What developers and tech leads should watch

For senior engineers and technical decision-makers, Hang Ten is another sign that AI-assisted development is moving from individual productivity into delivery-model redesign.

That shift changes the evaluation criteria. Asking whether an AI tool can autocomplete code is too narrow. Teams need to know whether AI can participate safely in the software lifecycle.

A serious enterprise setup needs:

• strong repository and environment isolation
• audit trails for AI-generated changes
• policy controls around data exposure
• integration with issue trackers, CI/CD, test suites, and code review
• model evaluation against internal coding standards
• rollback and incident response plans
• clear ownership for generated code

Security deserves special attention. Any system that reads proprietary code, touches internal APIs, or proposes production changes becomes part of the software supply chain. If access controls, logging, and approval workflows are weak, AI-assisted delivery can create new risk faster than it removes old work.

Hang Ten’s pitch is ambitious, and the target is real. The IT services model is vulnerable wherever work is repetitive, expensive, and buried under process. The harder test is whether Hang Ten can make AI agents useful inside messy enterprise systems without turning every engagement into bespoke consulting.

If it can, the pressure on traditional services firms will get harder to ignore.