Edition in partnership with

Welcome executives and professionals. Personal AI agents require organizations to draw the line between enterprise standardization and individual autonomy.

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:

• Personalized AI agents at work.

• The AI-first technology function.

• State of AI Report 2026.

• Coinbase’s AI agent cuts fraud.

• Transformation and technology.

• Insights for Executive+ members.

• Career opportunities & events.

Read time: 4 minutes.

Important note: The next edition of Enterprise AI Executive will be published on Wednesday, October 21, with new editions every Wednesday thereafter. We will no longer publish on Sundays.

MARKET & BEST PRACTICE INSIGHT

Personalized AI agents at work

Brief: Deloitte published an analysis of personal AI agents that employees build or bring to work, arguing that they can lift individual productivity and speed adoption but raise questions about security, cost, trust and control.

Breakdown:

• Almost half (49%) of workers using AI use unapproved tools, and senior leaders are 32% more likely to do so than other employees.

• Only one in five companies has mature governance for autonomous agents, yet three-quarters plan to deploy them within two years.

• Beyond enterprise-wide agents, firms can let employees build personal agents on approved platforms or bring their own (image above).

• Instead of banning personal agents, the report favours piloting them for individual or small-team use, away from sensitive data.

Why it’s important: Shadow AI reaches the top, with senior leaders more likely than other employees to use unapproved tools. When the people setting the rules work around them, banning personal agents won’t stop their use, only hide it. An approved route enables limits and clear oversight.

IN PARTNERSHIP WITH ELEVENLABS

The #1 text to speech model

Brief: Built on a new architecture, Eleven v4 is ElevenLabs' most emotive voice model, ranked #1 by Artificial Analysis. Eleven v4 Turbo brings the same research to low-latency use cases such as agents, at ~100ms median latency.

Hear the difference:

• Both interpret tone, pacing, and emotion for speech that feels performed

• Voice clones: Instant from 10 seconds of audio, Professional for top fidelity

• Long-form speaker consistency, IPA and multi-speaker dialogue

• Inline tags control delivery, emotion, pacing, reactions, SFX and style

Get started for free.

BEST PRACTICE INSIGHT

The AI-first technology function

Brief: BCG published a 23-page executive playbook on how agentic AI is moving the tech function from service provider to strategic partner, built on AI-first delivery, an agent-ready tech estate and closer C-suite partnership.

Breakdown:

• BCG says AI-first tech teams can speed software delivery 2-5x and lift tech productivity by 30-70%, with “power teams” at the top end.

• At one hyperscaler, six-week sprints raised power users of agentic coding tools from 5% to 50% of engineers, doubling productivity.

• Tech leaders are encouraged to build a vendor-neutral agent platform with an agent registry, cost gateway, memory and security built in.

• CIOs should partner with the CFO on agent economics, including a “synthetic P&L” for AI, and with the CHRO on human-agent teams.

Why it's important: Agents make software spend look more like a workforce budget. That makes the CIO a peer to the CFO and CHRO, not just a supplier to them. CEOs who keep tech in a service role will see AI managed as a cost line, not used to reinvent work and drive growth.

MARKET INSIGHT

State of AI Report 2026

Brief: Air Street Capital released its ninth annual State of AI Report, a 244-slide review spanning research, industry, politics and safety, with its industry findings showing that AI revenue and adoption are accelerating fast.

Breakdown:

• OpenAI and Anthropic’s combined revenue run rate reached about $105B by August 2026, up from roughly $30B in early January 2026.

• Top-quartile AI-native firms with $20M+ revenue grew at least 172% a year, versus 53% for those adding AI to software (image above).

• The cost of reaching a given level of AI performance has nearly halved every quarter since 2023, making it about 13x cheaper a year.

• Better tools, context and feedback lifted GLM-5.1’s success on 100 coding tasks from 52.5% to 65.5%, without changing the model.

Why it’s important: Enterprises can all rent similar frontier models, so the model is increasingly commoditized. The advantage is the scaffolding around it: tools, context and feedback. That added 13 points here, and it’s the part of your stack rivals find harder to copy.

AI-NATIVE PROFESSIONAL

How to refresh a forecast with ChatGPT

Brief: In this guide, learn how to give ChatGPT actuals, budget, forecasts, KPI data, thresholds, and owner notes, then ask it to rank drivers behind material movements and build a variance bridge in a spreadsheet you can edit.

Step-by-step:

1. Define the periods and versions to compare, such as forecast to actual, budget to actual, or current forecast vs prior forecast.

2. Attach actuals, budget, forecasts, KPI data, and owner notes, and set a materiality threshold, sign conventions, and finance lines.

3. Run the starter prompt and ask for the variance bridge as a spreadsheet you can edit, then check source breaks, signs, and drivers.

4. Continue in the same chat to resolve owner questions, confirm open reconciliations, and prepare clear variance commentary to review.

Best practice: Driver commentary should follow evidence. Use the follow-up prompt so ChatGPT labels hypotheses when files show a move but no cause.

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

CASE STUDY

Coinbase’s AI agent cuts fraud

Brief: Coinbase detailed how it built a fraud agent on Claude Opus to catch attacks its ML models missed when signals spread across transactions, cutting fraudulent transactions 30% and fraud value 22% in a live test.

Breakdown:

• Selective routing: the agent reviews only a subset of transactions after existing checks, balancing fraud detection, speed and cost.

• Structured context: each review combines the current transaction with summaries of recent account history and clear review guidance.

• Bounded authority: the agent returns a risk rating and can add a block, but cannot lift an existing one or execute a payment itself.

• Built-in reliability: retries, a fallback model and logs of every decision, model and timing keep it running and make it auditable.

Why it’s important: Coinbase didn’t replace its fraud stack. It added a narrow layer where existing technology was weakest, with projected savings around 3x the running cost. It shows how enterprises can leverage targeted agent capabilities to improve outcomes where current systems fall short.

Anthropic shared how Cresta built Conductor, an agent that builds other agents, and published a guide to reliable agent automations.

Thoughtworks explored how leaders should rethink roles, structure, decision rights and accountability as AI agents join their teams.

GLG published a 27-page report on closing the AI value gap, with five experts diagnosing where ROI breaks down in implementation.

AMD published a 12-page guide to building AI architecture that can evolve as models, technologies and business priorities change.

Deutsche Bank published a 20-page report on what superintelligence will feel like and when it is likely to arrive.

HandsOn published an 18-page whitepaper introducing its CLEAR framework, an approach to help adopt and scale enterprise AI.

Meta released Muse Wonder and Muse Wonder Full, with the full version ranking #2 on Artificial Analysis’ index, behind Opus 5.5.

Nvidia is reportedly exploring an acquisition or acqui-hire of Reflection AI, the $25B U.S. open model startup it already backs.

Anthropic detailed new defenses against rival labs copying Claude through distillation, saying they cut harvested data by 90-99%.

Microsoft unveiled Maia 300, its in-house AI chip promising 40% more AI output than leading GPUs per watt, powering Copilot by 2027.

US Justice Department charged an OpenAI engineer with allegedly hiding a recording device in an executive conference room at OpenAI.

Microsoft CEO Satya Nadella published an essay urging firms to guard AI models with an emergency brake and controls AI can’t touch.

CAREER OPPORTUNITIES

Siemens - Head of Data & AI

Citi - AI Adoption Senior Vice President

Cognizant - Frontier AI Principal

EVENTS

Gartner IT Symposium/Xpo™ - October 19-22, 2026

Wipro - Agentic AI Scaling - October 21, 2026

MIT - Leading in an AI World - November 03, 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.