Hang Ten Sells Outcomes

Hang Ten is not shipping a chatbot. It is trying to eat the systems-integrator line item.

On 16 September 2026, TechCrunch reported the former Infosys CEO Vishal Sikka's four-month-old company added $53 million to seed, five weeks after a $32 million first close. Hang Ten's own site now prints $85 million total. Xora (Temasek-backed) led the add-on; Mayfield and Aramco Ventures came back. Jerry Yang joined the board. Headcount is still about 20-25 people across the U.S., the Middle East, and Australia, with hiring planned in Europe and India.

The product they put under that raise is Hobie: an agentic platform with five named layers, plus a small forward-deployed engineering team, plus outcome-based pricing. No new foundation model. Skills packed for regulated work.

Hobie's five layers

Shoreline is the chat surface, including pipes into Claude and ChatGPT Desktop.

Shaper builds and refines agents by reading live systems, then loops evaluate, diagnose, fix, verify. Hang Ten says most of that loop is deterministic compilation, not free-form generation.

Lens is a versioned semantic model: datasets, metrics, relationships, a vocabulary you can edit without touching database views.

Reef queries existing systems in place. Nothing copied, nothing migrated. One model mapped onto many backends.

Beacon is audit: every answer traces to data and model version, fail-closed, role-aware, per-caller credentials, with HIPAA / SOX / NIST language on the site.

CTO Sanjay Rajagopalan told TechCrunch the in-house framework packages reusable AI skills for regulated industries. Customers or third parties still handle final sign-off. The public promise is a 10-fold improvement in cost, speed, or a mix, agreed before the work is scoped, not discovered in a QBR.

Where the hours actually go

Hang Ten's own diagram is the tell. When agents write the code, Build shrinks. Plan and Review grow, because the same people now handle more changes, and the recovered time is supposed to pay for running every control on every diff. That is an architecture claim: the bottleneck moves from typing to waiting on humans who can still say no.

They cite third-party ranges (METR 2025; GitHub/MIT Copilot studies; McKinsey State of AI 2024) for 26-55 percent productivity lift and a 2-5x cut in the customization tax. Those are industry studies, not Hang Ten customer RCTs. Ask for the program-level baseline before you treat 10x as a contract.

The work they list: SAP landscapes and migration, finance/HR/analytics (AI Hat), product software on an agentic SDLC, and CoE advisory. Named gravity: Siemens Gamesa (Vinod Philip), an energy supermajor via Aramco Ventures (Mahdi Aladel), plus healthcare/pharma case blurbs. Mayfield's Navin Chaddha frames the TAM as people-spend, not seat-spend: white-collar labor vs a much smaller enterprise software market, pricing moving from subscription to outcomes.

What to watch

A 25-person firm selling 10x on $10B-revenue accounts is a leverage story. The risk is the old SI risk with a new compiler: skills that only the vendor can run, data that never leaves the customer but logic that lives in Lens, and outcome contracts that hide who owns the failure when an agent's compiled query is wrong.

If you already run Cursor or Claude on the inner loop, Hang Ten is competing for the outer loop: the budgeted transformation program. That is a different buyer than a staff engineer. Treat Hobie as a control plane claim: semantic model, in-place queries, fail-closed access, and demand the Beacon traces in the RFP.

Sources

Hang Ten Systems, product and Hobie architecture. https://hangten.ai/

TechCrunch, Former Infosys chief's AI startup nabs another $53M, 16 September 2026. https://techcrunch.com/2026/09/16/former-infosys-chiefs-ai-startup-adds-50m-to-seed-weeks-after-initial-raise/

Mayfield, Rebuilding Enterprise Software for the Agentic Era: Hang Ten Systems. https://www.mayfield.com/rebuilding-enterprise-software-for-the-agentic-era-hang-ten-systems/