Skip to main content Skip to footer
  • "com.cts.aem.core.models.NavigationItem@5be9cd93" Careers
  • "com.cts.aem.core.models.NavigationItem@607a4df3" News
  • "com.cts.aem.core.models.NavigationItem@27adcec0" Events
  • "com.cts.aem.core.models.NavigationItem@9a56f02" Investors
AI Builder Brief - Issue 4

October 05, 2026

10 mins read


The AI Builder Brief - Issue 4

Welcome to The AI Builder Brief, your curated guide to the enterprise AI developments that matter. Every issue brings you the most important AI stories, research and reports, interpreted by Cognizant's AI leaders in plain terms.

In this issue: Should we slow AI down? In TIME this week, Cognizant's Ravi Kumar S, Babak Hodjat and Jeff DeMarrais argue that's the wrong question. Stig Martin Fiskå makes the case that Responsible AI is an engineering discipline, not an emergency brake. Amir Banifatemi explains what the new White House safety accord does, and doesn't, mean for enterprises. Babak picks out G5, which makes natural-language programming reliable by compiling it through a structured ontology. And the Cognizant AI Lab shares three new pieces of research: a robot that checks the law before it acts, a deterministic arbiter that keeps agent transactions under coded control, and a 75-million-parameter model that rivals far larger ones on ARC-AGI 1. The message for builders is simple: capability isn't the bottleneck anymore. Structure is. 

We hope you enjoy the read.

 

In TIME, Cognizant's Ravi Kumar S, Babak Hodjat and Jeff DeMarrais argue against a coordinated pause on frontier AI. They make the case that the future of AI lies in networks of specialized, governed agents rather than one all-powerful system.

EXECUTIVE'S TAKE

"Slowing the AI frontier won't make us safer. The path forward is to advance and control, with clear roles, bounded workflows and full accountability for what we deploy. Start where AI can be measured and governed, in workflows with structured inputs, clear outcomes, high volumes and fast feedback, and then build the controls in before you scale."

- Ravi Kumar S, CEO


A new platform makes natural language a reliable way to program, which could move software creation from specialist developers to the people who know the business best.

 

EXECUTIVE'S TAKE

"G5, from an MIT spin-out, lets people write software in plain English and compiles it to structured code and back, using what it calls "a self-learning, bi-directional compiler" built on an ontology graph. That means product owners, analysts and compliance leads can build and govern software themselves, and policies and guardrails are easier to enforce. Because the code can be translated back into plain language, the people who own the business logic can check that the software does what they intended. It's a strong example of this issue's theme, where adding structure makes AI more reliable and opens it up to more people."

- Babak Hodjat, Chief AI Officer


In a widely shared commentary, The Atlantic CEO Nicholas Thompson unpacked the recent high-profile researcher resignations at Anthropic and the internal warnings that the race to the frontier is pushing alignment and safety down the priority list.

 

EXECUTIVE'S TAKE

"The recent resignations at frontier AI labs are a wake-up call about putting speed ahead of safety. But the answer isn't panic. It's disciplined engineering. Models don't set their own goals. Humans define the objectives, build the harnesses and decide where capital goes. That's why Responsible AI is the core mission, not a sidebar. Sometimes that means drawing a hard line, but more often it means pointing AI at what matters most, from augmenting human expertise to building climate resilience and stronger public infrastructure. We set the boundaries, and we hold the steering wheel."

- Stig Martin Fiskå, Global Head of AI for Good


Six frontier labs, including OpenAI, Google and Anthropic, have signed a voluntary White House accord that builds safety controls around AI rather than slowing it down. It sets four layers of oversight: internal controls, an internal team to check them, an independent auditor and a board committee. Experts call it a positive start but say it will only work if the audits are genuinely independent.

EXECUTIVE'S TAKE

"This is a welcome step, and it fits a wider shift from debating whether to slow AI down to building the structures that keep it under control. Enterprises should be clear about its limits, though. The accord covers the companies that build frontier models, not the organizations that deploy them. A vendor signing it shows a commitment to a process, not proof that a model is safe for your use case. Every model still needs to be tested against your own risks, data and workflows. The accord also doesn't replace regulations such as the EU AI Act or standards such as ISO/IEC 42001 and the NIST AI Risk Management Framework, which remain the foundation of responsible deployment. The responsibility that matters most sits with the enterprise, at the point where AI meets real decisions."

- Amir Banifatemi, Chief Responsible AI Officer

From the Cognizant AI Lab

Our researchers have been putting this issue's theme into practice. These three projects show how structure makes AI more trustworthy and more efficient. One builds the law into how a robot plans, one keeps agent transactions under coded control, and one shows that a small model designed around the problem can rival systems many times its size.

The Legislated Planner checks a robot's planned actions against encoded laws before it moves, which made the robot about six times more likely to follow the rules.


Babak Hodjat shows how agents can negotiate freely while prices, budgets and settlement stay under coded control.


A 75-million-parameter model trained in about four hours on a single GPU rivals far larger reasoning models, because it was designed around the problem's structure.


Thanks for reading the latest edition of The AI Builder Brief. Subscribe to get the latest news sent directly to your inbox.


The AI Builder Brief curates the enterprise AI developments that matter and tells you why, in plain terms. Every issue is filtered and interpreted by Cognizant's AI leaders so you can lead with confidence, not catch up on weekends.


Subscribe on Substack

Get expert Cognizant executive advice straight to your inbox.


Past Issue