Generative AI Enterprise
Posts
Microsoft executives reveal best practices

Microsoft executives reveal best practices

Plus, AI in China, OpenAI's reasoning best practices, and more.

Lewis Walker
February 23, 2025

WELCOME, EXECUTIVES AND PROFESSIONALS.

Microsoft has a history of capitalizing on platform shifts, from client-server computing to the internet and cloud. But with the generative AI shift evolving faster than ever before, how can enterprises seize the latest opportunity?

Since the previous edition, we've reviewed hundreds of the latest insights on best practices, case studies, and innovation. Here’s the top 1%...

In today’s edition:

Microsoft executives reveal gen AI best practices.
Deloitte: Managing four categories of gen AI risks.
The state of AI models in China.
Sequoia in conversation with CEO Nikesh Arora.
OpenAI publishes reasoning best practices.
Transformation and technology in the news.
Career opportunities & events.

Read time: 4 minutes.

BEST PRACTICE INSIGHT & CASE STUDIES

Microsoft executives reveal gen AI best practices

Image source: Microsoft

Brief: Microsoft published The AI Decision Brief, a 38-slide guide for AI leaders on navigating gen AI. It covers the landscape, future potential, and best practice strategies, featuring executive perspectives, market studies, and real-world examples.

Breakdown:

AI capabilities are accelerating rapidly, doubling 4x faster than historical progress in client-server computing, the internet, and mobile/cloud.
AI usage and sentiment is trending upwards with leading “AI-rich” enterprises achieving 4.9x average ROI.
Microsoft emphasizes learning from AI-native startups, software firms, co-innovation labs, and research groups to help drive innovation.
Best practices in areas like strategy and data are highlighted as enterprises progress from exploring AI to scaling it for maximum value.
Brad Smith, Vice Chair and President, and Nathalie D’Hers, CVP Microsoft Digital, are among the many executives who share their perspectives.
Dentsu’s success in cutting time to media insights by 90% using Azure AI and Copilot is shared, alongside other enterprise case studies.

Why it’s important: Microsoft holds a strong position in the enterprise AI market, with 70% of Fortune 500 companies using Microsoft Copilot, its OpenAI partnership, and Azure's 25% share of the global cloud market. Leveraging experiences from its executives and customers can help turn AI ambitions into lasting value.

BEST PRACTICE INSIGHT

Deloitte: Managing four categories of gen AI risks

Image source: Deloitte

Brief: Deloitte released Part three of its Engineering in the Age of Generative AI series, exploring four categories of gen AI risks: enterprise, gen AI capabilities, adversarial AI, marketplace, and how leaders can manage them.

Breakdown:

Risks to enterprises include threats to operations and data, such as IP risks from the origins of training data.
Risks to gen AI capabilities include the potential for AI malfunctions and vulnerabilities like prompt injections and hallucinations.
Adversarial AI risks involve threats from malicious actors using gen AI for harmful purposes, such as phishing attacks.
Marketplace risks involve economic, legal, and competitive pressures, including regulatory uncertainties and data center capacity.
The article outlines 13 risk mitigations, from implementing digital asset management to creating data center capacity via low Earth orbit.
Deloitte emphasizes the need for a human-in-the-loop approach and secure-by-design principles across the software development lifecycle.

Why it’s important: While gen AI offers significant opportunities to improve enterprise products and practices, it also introduces risks. No single solution can address them all. Deloitte's framework can help cyber and risk leaders assess risks and tailor multiple mitigation strategies based on their exposure.

MARKET INSIGHT

The state of AI models in China

Image source: Artificial Analysis

Brief: Artificial Analysis published a 14-slide report on the state of AI in China charting the rise of China’s top AI companies, mapping the Chinese AI ecosystem and comparing intelligence to leading US models.

Breakdown:

Chinese AI labs have narrowed the gap with U.S. labs, with DeepSeek’s January-released R1 model now rivaling OpenAI o1-level intelligence.
Alibaba led China's AI models until DeepSeek pulled ahead in late 2024, postioning itself as China's top AI lab.
In early 2025, Chinese labs like MoonShot, Tencent, Zhipu, and Baichuan released new competitive AI models, signaling strong ecosystem growth.
In the USA, Google and Anthropic have narrowed the gap on OpenAI. Gemini 2.0 Flash now outperforms Claude 3.5 Sonnet, but o3 still leads.
This report predates Grok 3, ranked second to OpenAI's o3. The trajectory of Grok 3's intelligence improvements is one to watch closely.

Why it’s important: The analysis highlights the convergence and concentration of frontier model intelligence in the USA and China. Understanding this trajectory, along with other factors like cost, customization, and regulatory alignment, can help enterprises leverage the right models for their requirements.

EXECUTIVE INSIGHT

Sequoia in conversation with CEO Nikesh Arora

Image source: Sequoia

Brief: Sequoia spoke with Palo Alto Networks CEO Nikesh Arora, who leads the world’s largest cybersecurity company. He emphasized the balance between innovation and responsibility, especially in AI security and enterprise adoption.

Breakdown:

Mission-critical enterprise apps need precision models with high-quality, industry-specific data. Proprietary data and domain expertise is key.
Enterprises should implement robust guardrails, monitoring, and controls before deploying in production, including human oversight of critical systems.
Moving fast in AI requires clear vision and detailed execution plans. Leaders must set a North Star, ensure resources, and remove obstacles.
Enterprise AI applications cannot afford the error tolerance of consumer models. Mistakes can disrupt infrastructure or enable bad actors.
Companies should experiment aggressively in contained environments. Managing risks enables companies to capture AI’s benefits.
The aspect of preparing data infrastructure remains a significant hurdle, but Nikesh cautions, “It would be irresponsible not to experiment.”

Why it’s important: Palo Alto Networks' revenue has grown over 185% since Nikesh Arora became CEO in 2018, excelling in identifying cybersecurity vulnerabilities faster than adversaries. Arora’s insights are valuable for success in AI security and broader enterprise transformation.

BEST PRACTICE INSIGHT

OpenAI publishes reasoning best practices

Image source: OpenAI

Brief: OpenAI shared reasoning best practices, explaining their differences from non-reasoning models, when to use them, and how to prompt them for optimal results. Timely guidance for professionals in a fast-changing market.

Breakdown:

OpenAI offers two model types: reasoning models (e.g. o1 and o3-mini) and GPT models (e.g. GPT-4o). These model families behave differently.
If speed and cost matter most and your use case involves straightforward, well-defined tasks, GPT models are the best choice for efficiency.
If accuracy and reliability matter most and you’re tackling a complex, multi-step problem, o-series models are the best choice for your needs.
Most AI workflows combine both models: o-series for agentic planning and decision-making, and GPT models for efficient task execution.
In the image above, GPT-4o triages customer order details, detects issues, and sends data to o3-mini, which determines return viability.
The guide outlines when to use reasoning models, including 7 patterns like needle in a haystack and multi-step planning, plus tips on effective prompting.

Why it’s important: The rapid proliferation of models has led to increasingly complex OpenAI offerings. This guide provides clarity on when to leverage o-series models ("the planners") and GPT models ("the workhorses") for cost-effective outcomes. OpenAI’s GPT-5, coming in “months”, will unify both series.

BCG published a 24-slide playbook to help executives capitalize on AI opportunities in R&D, expanding its series of AI transformation guides, covering sectors from finance to HR. BCG also shared insights on the CIO's role in AI value creation.

Bain Capital Ventures surveyed its CFO Advisory Board, a community of 50 CFOs, revealing that 79% plan to increase AI budgets in 2025. 94% see strong potential for gen AI in finance, though adoption lags with 71% not using it yet.

OWASP released a 48-page paper on agentic AI threats and mitigations, covering a reference architecture and a comprehensive model addressing threats from memory poisoning to rogue agents, with guidance on identification and mitigation.

AWS shared a case study on Formula 1 using gen AI to speed up race issue resolution. The team built a root cause analysis assistant with Amazon Bedrock, reducing issue resolution from weeks to minutes.

Forrester, in partnership with AWS, published a 21-page report Mining the Generative AI Goldrush. The survey of 657 software leaders found many rushing to deploy AI, which may hinder long-term success without adequate preparation.

OpenAI COO Brad Lightcap revealed the company now has 400M weekly users and 2M enterprise customers, with developer usage doubling in six months, demonstrating strong adoption across consumer and enterprise markets.

Microsoft shared a 57-slide presentation on technical gen AI best practices, covering model selection, performance, lifecycle management, RAG, agents and Azure AI Agent Service within Azure AI Foundry.

xAI and Elon Musk launched Grok-3, claiming it’s the "smartest AI on Earth," outperforming GPT-4o, Claude 3.5 Sonnet, and Gemini-2 Pro. With Palantir adopting Grok, xAI plans to release an enterprise API for Grok 3 and Grok 3 Mini soon.

DeepSeek announced it will open-source five repositories next week. While specifics are unannounced, the move seeks to introduce building blocks that are "documented, deployed, and battle-tested in production."

Meta revealed a new initiative to develop AI, hardware, and software platforms for humanoid robots, positioning itself as the foundational engine powering the market rather than focusing on consumer products.