Edition in partnership with

Welcome executives and professionals. Years have been spent explaining machine reasoning to earn people's trust. The harder challenge now is teaching AI to understand and apply human judgment.

Since the previous edition, we have reviewed hundreds of the latest insights in agentic and generative AI, spanning best practices, case studies, market dynamics, and innovations.

This briefing outlines what is driving material value — and why it’s important.

In today’s briefing:

• McKinsey open-sources design system.

• Four questions for every CIO.

• The AI-first chief operating officer.

• From AI activity to enterprise advantage.

• Transformation and technology.

• Insights for Executive+ members.

• Career opportunities & events.

Read time: 4 minutes.

ACCELERATOR

McKinsey open-sources design system

Brief: QuantumBlack, AI by McKinsey open-sourced its design system, rebuilt so that AI agents as well as designers/developers can use it. As building gets easier, judgement about what to build becomes the real differentiator.

Breakdown:

• QuantumBlack's design system (QBDS) worked well for designers and developers, but it wasn’t set up for AI-assisted development.

• The previous system held components, design tokens and documentation. QBDS adds a design-code link, guidelines, intent and agent skills.

• A prompt loads those skills, so agents build with the system's rules and components rather than guessing at conventions.

• Custom components in npm packages are hard for models to read, so the system was rebuilt on shadcn and Tailwind foundations.

Why it’s important: Every token, guideline and connection between design and code should be explicit: the less a model guesses, the fewer iterations and the lower cost to production. A well-structured design system is among the best investments a team can make. McKinsey’s is now on GitHub.

IN PARTNERSHIP WITH TELEPORT

Simple prompts beat complicated harnesses

Brief: Teleport put 13 engineers on three months of "pressure washing," pointing frontier models at its own codebase to find security bugs. The result: nearly twice the high severity vulnerabilities they logged in 2024 and 2025 combined.

By Rob Picard, Teleport. 5 min read. In this post:

• See why Conclave, a multi stage harness, did not beat a prompt

• Copy the CTF style prompt behind their bug finding work

• Watch human triage become the bottleneck as volume rises

• Weigh three months against 1 to 2 years of “pre-AI” hardening

Read the post.

BEST PRACTICE INSIGHT

Four questions for every CIO

Brief: BCG Platinion detailed how CEOs are betting on AI to reshape the enterprise. As CIO, are you ready to lead a boardroom conversation about the architecture that will make that ambition possible?

Breakdown:

• The AI reshape is a board-level obligation. CEOs are backing multi-year transformation plans and asking the CIO to determine what is possible.

• Reshapes stall because core data is fragmented, systems were built for human clicks, and agent identity models are inadequate.

• Key decisions trade compliance and predictability for speed and flexibility, settling whether a curated or open architecture fits.

• Four questions decide whether a reshape scales: one version of core data, a vendor strategy, agent identity, agent-ready systems (image above).

Why it's important: Answering these is not simply an IT exercise. CIOs now sit alongside the CFO and COO as partners accountable for growth. The right architecture strategy creates first-mover advantage; the wrong one commits capital that belonged elsewhere in the organization.

BEST PRACTICE INSIGHT

The AI-first chief operating officer

Brief: BCG outlined what CEOs should look for in an AI-first chief operating officer. Capturing productivity and cost gains needs a COO who pairs timeless operational strengths with a clear vision of what they want AI to do.

Breakdown:

• BCG estimates an end-to-end transformation could raise industrial operations productivity by more than 30% within two to three years.

• The difference between success and failure will rest largely on how well their chief operating officer (COO) develops an AI-first operating model.

• The COO's core mandate holds: cost, delivery, quality, innovation and time-to-market. How the COO delivers on it changes considerably.

• The task is reinvention from first principles, not automating today's model, and shifting from reacting to problems to acting on early signals.

Why it’s important: CEOs need bold executives who can challenge assumptions, reimagine how the business works, and keep people focused through relentless change. The question is which enduring COO strengths matter even more now, and which new capabilities the role demands.

AI-NATIVE PROFESSIONAL

How to produce reports with ChatGPT Work

Brief: In this guide, you'll learn how to ask ChatGPT Work to inspect raw data files, profile a merge before joining, explore the question with charts, and ship a report with the answer first and the caveats stated plainly.

Step-by-step:

1. Attach the CSVs or Excel workbooks, or name an approved Google Sheet, and ask ChatGPT to inventory what each file contains first.

2. Leverage the starter prompt. ChatGPT then profiles candidate join keys, duplicates and unmatched records before it merges anything.

3. Ask for charts tied to the original question, then an interpretable baseline model that states its assumptions and uncertainty plainly.

4. Always keep source files unchanged, have it name every file it creates, and flag any data issue that could change the conclusion.

Best practice: Leverage the follow-up prompts to package the analysis for stakeholders, asking for a memo or spreadsheet depending on who will read it.

For the full guide, including prompts, upgrade to Executive+ or The Boardroom.

BEST PRACTICE INSIGHT

From AI activity to enterprise advantage

Brief: Kearney examined what to do with the AI experimentation already under way. Copilots and local pilots are now more common, but they can create the appearance of progress while the operating model underneath stays unchanged.

Breakdown:

• Three misconceptions hold enterprise AI back: more use cases mean maturity, a tool is transformation, functions maximise value alone.

• Ask where better or faster intelligence could materially change performance, then set a few priorities, not another use-case catalogue.

• Then map priority workflows end to end, and decide what should be automated, what augmented, and what orchestrated across teams.

• Establish governance for scaling, not just approving: who owns the workflow, and where human judgment stays mandatory.

Why it’s important: Enterprise AI is not a contest to build the longest list of use cases. It is a design challenge: deciding where intelligence sits and how humans and machines share execution. Get that right and AI moves beyond an innovation activity to become a new way for the enterprise to operate.

BCG examined the CIO's role as business users build their own CRM agents, in an eight-page guide to owning four foundations (e.g. data architecture).

McKinsey published a blueprint for scaling agents, arguing enterprises deploy faster than they redesign work underneath.

Infosys detailed how enterprises can contain agent sprawl with automated governance checks as creation moves beyond the IT function.

Ropes & Gray examined a new category of shareholder litigation targeting the business decisions behind adoption, not the AI technology itself.

OpenAI profiled Fyxer, whose executive assistant drafts replies in each user's voice after training on 500,000 hours of workflows.

Sapphire Ventures released a market memo on how cheaper, faster, more autonomous attacks are reshaping cybersecurity priorities.

Salesforce introduced Koa, an in-house reasoning model for its sales and support agents, built on Nvidia's open Nemotron 3 Super.

Nvidia, Palantir, and Booz Allen Hamilton are reportedly limiting Fable on sensitive work over Anthropic's 30-day log retention.

U.S. President Donald Trump dismissed Dario Amodei's call to slow frontier AI, with China also labeling the same plan as "fear-mongering."

Chinese AI researchers published a roadmap of five levels of recursive self-improvement, ending with AI that builds its successors.

OpenAI CEO Sam Altman ruled out an OpenAI IPO in 2026, calling this an "ill-advised moment" to go public, given AI safety concerns.

Elon Musk proposed that the top US labs and leading Chinese companies run each other's safety tests before any frontier model ships.

CAREER OPPORTUNITIES

Fidelity - AI Strategy Director

Anthropic - Head of Regulated Industries

OpenAI - Head of Analyst Relations

EVENTS

McKinsey - Agentic AI Economics - September 22, 2026

Deloitte - Scaling Gemini Enterprise - October 1, 2026

AWS - Agent Factory Build - October 1, 2026

Reach enterprise AI decision-makers:

• 66% of readers are C-level executives or VP and Director-level leaders.

• 63.2% of the audience is based in the U.S., EU, UK, ANZ, and Singapore.

• Read by leaders at Microsoft, Deloitte, the Fortune 500, and more.

Guaranteed impression and custom sponsorship packages available, with post-send performance reporting.

Conceived as a practical communication for executives Lewis Walker has worked with, this briefing has become a trusted resource for thousands of senior decision-makers shaping the future of enterprise AI.

We welcome your feedback.