Jul 22, 2026, 2:33 PM ETFireside Chat, Interview

AI Is Only as Fast as Your Slowest Process | Jyoti Bansal, Harness | RAISE Summit 2026

Moved from India to Silicon Valley 25 years ago, launching a career in the developer tool chain space.
Founded AppDynamics in the software observability sector, selling it to Cisco for approximately $4 billion one day prior to an anticipated IPO in 2017.
Launched Harness in 2017 to address the "downstream" bottlenecks in software delivery, specifically the 60-70% of engineering time spent on testing, security, compliance, and deployment rather than coding.
Positioned Harness as an AI-driven platform for CI/CD and software delivery, predating the current Generative AI boom.

Generative AI has increased code production volume by 2x–3x, yet actual shipped features to consumers have only increased by approximately 10%.
This disparity creates a critical bottleneck where downstream processes (testing, verification, security) cannot scale to match the velocity of AI-generated code.
Harness claims that automating these downstream processes allows companies to ship 2x more code, aligning delivery speed with generation speed.
Internal data from Harness shows the company has doubled its own software production capabilities after implementing these automation strategies.

Harness automates the simulation of human judgment in testing, code reviews, security checks, and compliance verification.
Implements a "risk-based autonomy" model where AI handles low-risk tasks automatically, while high-risk scenarios involving quality, security, or cost escalate to human approval.
Utilizes progressive rollouts (e.g., releasing to 1% of consumers) to automatically verify code functionality before broader deployment or triggering rollbacks if issues arise.
Targets high-stakes sectors requiring rigorous safety checks, such as banking and aviation, where a single bug can cause significant disruption.

Deployed by 8 out of the top 10 US banks, including JP Morgan Chase (40,000 software engineers) and Citibank (20,000 developers).
Adopted by major enterprises like United Airlines, which utilizes the platform for its 8,000 software developers to automate code shipping.
Customers cite the ability to safely integrate AI into legacy environments and large-scale legacy mainframes as a primary value proposition.

Addresses the industry-wide challenge of "token maxing" and rising compute costs by providing tools to tie token expenditure directly to business outcomes (ROI).
Offers internal instrumentation that tracks token usage per specific task (e.g., cost per customer support ticket fixed) to eliminate waste.
Advises developers on model selection efficiency, such as avoiding using frontier models for simple tasks to prevent excessive costs.
Shifts the narrative of AI from an experimental cost to a tracked major operational expense requiring measurable return on investment.

The adoption gap between the US and other regions (Europe, Middle East, Asia) has narrowed from 3–8 years in previous eras to approximately 2 years.
US leadership in adoption remains, but global urgency is driven by bottom-up pressure from employees and developers demanding AI integration.
Regulatory constraints and government restrictions on data centers or models are viewed as catalysts for innovation, driving growth in open-source models and regional infrastructure (e.g., data centers in France).

Bansal expresses concern not about bottlenecks becoming unsolvable, but about potential excesses in compute, power, and hardware.
Suggests that global restraints on AI resources will force the development of more efficient systems and localized infrastructure.
Advises educators and parents to prioritize creativity over rote programming skills, warning that AI can homogenize thought processes if not counteracted by "out-of-the-box" thinking.
Recommends teaching children to leverage AI without blindly following its output, focusing instead on artistic and innovative problem-solving.