Generative AI Enterprise
Posts
McKinsey: Now leaders must step up

McKinsey: Now leaders must step up

Plus, CEOs react to DeepSeek, o3-mini, and more.

Lewis Walker
February 02, 2025

WELCOME, EXECUTIVES AND PROFESSIONALS.

McKinsey finds that the biggest barrier to scaling AI isn’t employees. It's leaders who aren’t steering fast enough. Meanwhile, US-China competition triggers the democratization of top reasoning models. CEOs react.

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:

McKinsey: Now leaders must step up.
CEOs react to DeepSeek's efficiency innovation.
OpenAI releases o3-mini reasoning models.
Google’s guide: From prototype to production.
Transformation and technology in the news.
Career opportunities & events.

Read time: 4 minutes.

MARKET INSIGHT

McKinsey: Now leaders must step up

Image source: McKinsey & Company

Brief: McKinsey’s 47-page report finds the biggest barrier to scaling AI is not employees, but leaders who aren't steering quickly enough. Inspired by Reid Hoffman’s new book 'Superagency,' it explores amplifying human potential with AI.

Breakdown:

Employees are more ready for change than leaders imagine. 3x more employees use gen AI for a third or more of their work than leaders think. 70% believe it’ll change 30% or more of their work within 2 years.
Companies need to move fast. Employees trust leaders to balance speed with safety. 47% of C-suite say gen AI is moving too slowly, despite 69% having invested over a year ago.
Employees are 1.3x more likely to trust their own companies to deploy gen AI right than other institutions.
Almost all companies invest in AI, but just 1 percent believe they are at maturity. 92% plan to invest more over the next 3 years.
Leaders need to recognize their role in driving gen AI transformation. Executives are 2.4x more likely to cite employee readiness as a barrier than leadership.
48% of employees say training is key for gen AI adoption, but nearly half feel they’re receiving moderate or no support.

Why it’s important: AI's rapid technological progress in the past two years is stunning. While some see it as a risk to humanity, what if we focused more on what could go right? Leaders may increasingly realize we're poised for a new era of productivity, innovation, and progress. The real risk? Thinking too small.

EXECUTIVE INSIGHT

CEOs react to DeepSeek's efficiency innovation

Image source: Social Capital

Brief: DeepSeek made headlines last week, sparking media coverage and debate. Enterprises have been assessing the implications of efficiency innovations in the Chinese company's R1 open-source reasoning model.

Breakdown:

Satya Nadella, Microsoft CEO, referenced the "Jevons Paradox," noting that greater efficiency and accessibility drive higher AI demand. R1 is now available on Azure AI Foundry.
Anthropic's CEO stated that DeepSeek's model “matches the performance of U.S. models 7-10 months older, for a good deal less cost”.
He emphasized that efficiency innovations and scale are both key, noting that superintelligent AI will still need millions of chips and billions in investment.
Palmer Luckey, Oculus VR founder, dismissed the widely reported $5-6M DeepSeek budget claim as inaccurate. Dylan Patel pointed out it only covers the final training run, excluding billions in CapEx and R&D costs.
Frank Meehan, an OpenAI investor, posted that regardless of the exact cost it’s cheaper, stressing, "Never underestimate China."
Zuckerberg doubled down on the importance of Meta's open-source strategy. IBM's CEO and Chairman Arvind Krishna stated, "Technology becomes truly transformative when it becomes more affordable."

Why it’s important: High performance at lower costs, driven by shared open-source progress and closed-source pricing adjustments, will expand the viability of AI-driven transformation opportunities for enterprises but increase compliance risks if models misalign with policies or regulations.

INNOVATION INSIGHT

OpenAI releases o3-mini reasoning models

Image source: Humanity’s last exam benchmark

Brief: OpenAI released o3-mini and o3-mini-high, its newest, most cost-efficient reasoning models. This early release, along with DeepSeek, marks significant step in democratizing access to increasingly high-end AI models.

Breakdown:

o3-mini is tailored for programming, math, and science, with function calling, structured outputs, and multi-level reasoning (low, medium, high).
The model delivers 39% fewer major errors on complex tasks and 24% faster responses than o1-mini. Outperforms DeepSeek R1 in reasoning.
Available free in both ChatGPT and the API, with higher query limits for Plus and Team, and Enterprise access expected in February (exact date TBC).
Priced at $1.10 per million input and $4.40 per million output tokens, versus DeepSeek R1's $0.55 input and $2.19 output.
As for o3-mini-high, it delivers higher-intelligence responses, ranking 200 Elo points higher than o1 on Codeforces coding tasks.
Pro users will have unlimited access to both o3-mini and o3-mini-high.

Why it’s important: Two weeks ago, top reasoning AI models were subscription-only. Now, free options are available via Microsoft Copilot (o1), ChatGPT (o3-mini), Google AI Studio (Gemini Flash Thinking), and DeepSeek (R1). However, the best models (o1-pro & o3-mini-high) still require a subscription.

BEST PRACTICE INSIGHT

Google’s guide: From prototype to production

Image source: Google Cloud

Brief: Google’s 36-slide guide provides a starting point for enterprises to take gen AI from prototype to production. Drawing on decades of experience operationalizing AI, it covers setting AI objectives, selecting the right models, evaluation, and more.

Breakdown:

More than 60% of enterprises are now actively using gen AI in production. In the past year alone, Gemini API usage on Vertex AI has surged 36x.
Driving value with gen AI requires defining business problems, prioritizing key use cases, and developing a comprehensive AI strategy.
The right platform matters. Invest in an AI platform, not just models. Some use cases may require multiple models to balance performance and cost.
You can’t improve what you don’t measure. Ensuring gen AI model reliability and accuracy is a major hurdle enterprises have to overcome.
Responsible AI is essential. Governance should be embedded from the start to ensure secure deployment and use across enterprises.

Why it’s important: Gen AI has massive potential, but realizing it takes careful execution. Google’s guide helps companies clarify objectives, select models, measure performance, and maintain AI in production, ensuring AI adoption is strategic, scalable, and impactful.

The International AI Safety Report was published, a 298-page effort produced by AI pioneer Yoshua Bengio and 96 AI experts to establish a shared understanding of AI risks and guide decision-makers on safe global adoption.

BCG released an 18-slide report finding that telcos' AI maturity compares well to other industries, but they lag in investment. It highlights how they are adapting to new trends and challenges, focusing on areas where telcos should prioritize their efforts.

McKinsey shared an article on how gen AI can enhance biopharma operations, from AI-assisted supervision on shop floors to smart deviation management, accelerating product development, and AI copilots for supply chain optimization.

AWS shared an article on gen AI operating models, detailing the impact of decentralized, centralized, and federated patterns on data access, risk management, reusable agents, compliance, architecture, and more.

Deloitte Legal published a 10-slide report on its 2025 AI expectations for in-house legal teams, spanning the transition to delivering real benefits and the evolving role and skillsets of legal professionals.

Microsoft AI CEO Mustafa Suleyman announced that ‘Think Deeper’ is now free for all Copilot users, integrating with OpenAI’s o1 reasoning model for enhanced insights.

The U.S. Copyright Office released a new report ruling that AI-generated outputs alone cannot receive copyright protection while preserving rights for human creators who use AI as a tool.

OpenAI is reportedly in talks to raise up to $40B, potentially more than doubling its $340B valuation from late 2024.

Italy banned the Chinese AI app DeepSeek over concerns about data sharing, storage, and GDPR compliance. Ireland, Belgium and others have launched investigations.

Meta CEO Mark Zuckerberg announced a $60-65B investment plan for 2025, focused on AI infrastructure to establish Meta AI as the leading assistant and Llama 4 as the state-of-the-art model in the industry.