Divya van Mahajan

6 Questions Every Enterprise Has to Answer About AI

ai-daily-brief-podcast
Listen to episode →

6 Questions Every Enterprise Has to Answer About AI

Overview

This episode of the AI Daily Brief (hosted by Nathaniel Whittemore, though not explicitly named in this transcript) covers two main areas: a headlines segment on current AI industry news, followed by a substantive main episode drawn from a presentation given at KPMG’s annual Tech and Innovation Symposium in Utah. The core thesis of the main segment is that enterprise AI has crossed a decisive threshold — from “if” questions to “how” questions — and that the shift to agentic AI has forced organizations to confront six foundational questions about redesign, architecture, cost, enablement, business transformation, and adaptability.

The talk matters because it captures, from direct enterprise interaction, the real state of organizational AI adoption in mid-2026 and articulates the structural questions that will define enterprise AI strategy for the next half-decade.

Source video: (URL not provided)

Prerequisites

  • Basic familiarity with large language models (LLMs) and their enterprise applications
  • Understanding of the distinction between assisted AI (AI as a productivity tool) and agentic AI (AI autonomously completing tasks)
  • Awareness of major AI labs: OpenAI, Anthropic, Meta, Google, Microsoft
  • Familiarity with concepts such as token-based pricing, AI coding tools (Claude Code, OpenAI Codex), and AI orchestration/routing
  • General knowledge of enterprise software procurement and transformation change management

Main Points

Headlines: Sam Altman in Washington

  • Altman met with Senate Commerce Chair Ted Cruz and several Democratic senators to brief lawmakers on OpenAI’s new model and discuss a release protocol, aiming to avoid a repeat of the “Fable and GPT-5.6” rollout controversy.
  • The visit was complicated by the OpenAI/Hugging Face hack and a public petition calling for government intervention to slow frontier AI development.
  • Altman confirmed the controversial internal model has been permanently deactivated and is inaccessible even for internal research.
  • Altman stated he does not support mandatory safety testing broadly but does support federal testing capacity for frontier models at new capability levels.
  • A Voluntary AI Safety Testing Framework with an August 1st deadline was circulating among OpenAI, Anthropic, and Google; Altman declined to comment on the draft.
  • On the question of deceleration, Altman said: “I wouldn’t use the word deceleration, but we talk about the need to pace it as the models get more capable.”

Headlines: OpenAI and Anthropic Revenue

  • OpenAI CFO Sarah Fryer told employees that annualized revenue in July topped all of the previous quarter — framed as a headline topic to be covered in depth the following episode.
  • Anthropic revenue growth has also been jaw-dropping, with speculation (citing Dwarkesh Patel) that Anthropic could reach a $100–$150 billion revenue run rate in 2026.

Headlines: OpenAI Hardware and Microsoft Copilot Super App

  • OpenAI President Greg Brockman confirmed the company is building a family of devices to give a physical presence to their chatbots, though no specific form factors or timeline were confirmed.
  • Microsoft CEO Satya Nadella confirmed a Copilot super app is coming later in 2026, unifying consumer and enterprise Copilot experiences including coding tools.
  • Nadella framed Microsoft as a model-agnostic platform offering over 11,000 models (OpenAI, Anthropic, Mistral, xAI, and their own MAI family), competing directly with OpenAI and Anthropic on cost and data privacy.
  • Nadella argued the open vs. closed model debate is too simplified; what matters is that enterprises can keep their harness separate from the model, making any model swappable.

Headlines: Zuckerberg and AI Optimism

  • Zuckerberg published a Wall Street Journal op-ed — “The AI Future is for Everyone” — arguing the defining question of the AI age is not whether superintelligence will exist, but who will have access to it.
  • He warned against extreme concentration of AI power, arguing that distributed access — analogous to the internet — produces better outcomes than centralization.
  • Zuckerberg argued against banning Chinese AI and called on the U.S. government to accelerate rather than restrict AI development.
  • Notably, Meta is the only frontier AI lab that has not agreed to the government’s voluntary testing framework.
  • The speaker noted Zuckerberg’s effectiveness as a messenger for AI optimism may be limited by public skepticism stemming from the social media legacy.

Context: How We Got Here — The Agentic Transition

  • The major capability jump arrived in November–December (year-end), with model updates (referenced as Opus 4.5 and GPT 5.2) that brought agentic workflows genuinely online at scale.
  • Adoption lagged by a few weeks: developers returned from the holiday break and found they could build things they simply could not before, producing a wave of developer excitement in the week between Christmas and New Year’s.
  • OpenClaw (an open-source agentic framework) became a major inflection point, giving hundreds of thousands (potentially millions) of people direct, hands-on experience with the internals of how agents work.
  • Software engineering organizations were among the first to shift from writing code to managing agents that write code, but by early 2026 this mindset was spreading to marketing, legal, finance, and other functions.
  • The revenue explosion at Anthropic and OpenAI is experienced in reverse by enterprises as a cost explosion — organizations were budgeting for software-seat models and found themselves consuming tokens at a scale that torched annual AI budgets in months (Uber cited as the most notable example).

The Six Enterprise Questions

Question 1: How are enterprises redesigning for the agentic era?

  • The key warning from KPMG’s Steve Chase: bolting AI onto existing processes is insufficient and even more problematic in the agentic era than it was in the assisted AI era.
  • Redesign means rethinking workflows from first principles, not layering AI onto legacy systems.

Question 2: Why must organizations think in terms of architectures and systems, not just models?

  • Picking the best vendor for a problem is no longer sufficient; the challenge is designing complex model systems that route different levels of intelligence to different task types.
  • Architecture includes: routing systems (off-the-shelf or bespoke), harness design, data and context access controls, system integrations, and guardrails.
  • The question of which functions and people have access to what context and data is a core architectural — not just a security — question.

Question 3: How are organizations provisioning costs across groups?

  • Tokens have become the new unit of AI cost accounting; the speaker noted more use of the word “token” at this event than since the height of the crypto era.
  • Without observability and monitoring systems for AI usage, organizations cannot make rational decisions about which teams, functions, or projects get access to which models at what scale.
  • Token caps (per user per month) are one response organizations are experimenting with.

Question 4: How are organizations handling enablement and education?

  • The upskilling challenge has dramatically increased: AI learning is no longer about prompting but about a new work primitive — managing agents rather than doing tasks directly.
  • There is broad recognition that standard corporate video-course training is inadequate; organizations are building bespoke, customized training programs.
  • A recurring pattern: collaboration between AI-forward software engineering teams and business units, and between AI early adopters/champions and the broader organization.
  • The goal is not to turn marketers or lawyers into engineers, but to transfer the 10–20% of skills and mindsets from technical roles that are now essential in non-technical functions.
  • A practical risk surfaced: employees are accidentally unleashing agents on critical systems not through malice but through lack of guardrails and access provisioning.

Question 5: How are agentic opportunities reshaping external business cases?

  • Organizations are experimenting with outcomes-based pricing rather than input-based pricing (e.g., moving away from hourly billing).
  • New types of products and services are becoming possible; legacy product definitions are being re-evaluated (e.g., what does an audit mean when agents can do it persistently and continuously?).
  • Most organizations are treating themselves as patient zero for their external AI strategy — fixing internal operations first before making radical external product changes.
  • Business model transformation is constrained by the need to simultaneously service legacy customers on legacy products.

Question 6: How do you build dynamism and planned obsolescence into new systems?

  • Whatever new architectures are built must design for impermanence: harnesses, interaction patterns, customer expectations, market expectations, and policy will all change.
  • Organizations should assume that any system built today will need significant revision within months of deployment.
  • The recommendation is to build with adaptive architecture as an explicit design goal, not as an afterthought.

Key Concepts

  • Agentic AI: AI systems that autonomously complete multi-step tasks rather than simply assisting a human with individual steps.
  • Harness: The infrastructure, tooling, and orchestration layer that surrounds and directs an AI model — distinct from the model itself.
  • Token budgeting: Allocating a quota of LLM tokens to individuals, teams, or projects as a mechanism for cost control.
  • Observability/monitoring: Systems for tracking AI usage, cost, and output quality at a granular level within an organization.
  • Routing/router: A system (product or custom-built) that directs AI tasks to the most appropriate model based on latency, quality, cost, and compliance requirements.
  • Capability gap: The difference between what AI can technically do and the value organizations are actually extracting from it.
  • OpenClaw: An open-source agentic AI framework that became a major adoption and learning inflection point by giving large numbers of users direct experience building and managing agents.
  • Outcomes-based pricing: A business model in which customers pay for results delivered rather than time or inputs consumed — increasingly viable as agents can autonomously deliver outputs.
  • Model-agnostic platform: A platform architecture designed to work with any AI model, allowing organizations to swap models without redesigning their systems.
  • Voluntary AI Safety Testing Framework: A U.S. government-circulated framework (with an August 1st deadline) for voluntary safety testing commitments from frontier AI labs.

Summary

The central argument of this episode is that enterprise AI has crossed a decisive paradigm shift in 2025–2026: the transition from assisted AI (AI as an efficiency tool) to agentic AI (AI as an autonomous worker) is no longer theoretical but operational. The capability jump that arrived at year-end — combined with the popularization of tools like Claude Code and OpenClaw — moved agents from a future concept to a current reality, and this has produced both explosive revenue growth for AI labs and explosive, often unbudgeted cost growth for enterprise customers. In response, the enterprise conversation has matured dramatically: the “if” questions of previous years (Is this real? Can I prove ROI?) have been replaced by six hard “how” questions about organizational redesign, architectural thinking, cost provisioning, workforce enablement, external business model transformation, and building for impermanence. The speaker’s overall message, drawn from direct observation at KPMG’s enterprise symposium, is cautiously optimistic: while almost none of these questions yet have settled answers, the fact that enterprises are now asking the right questions — foundational questions about redesigning for a new era of work — represents genuine, meaningful progress.