AI Transformation Brief – August 2025

On July 23, 2025, the US administration released America’s AI Action Plan, a 24- page policy that highlights the following three main initiatives:

• Accelerating AI innovation: Promotes a deregulated environment with respect to AI and open-source technology.
• Building American AI infrastructure: Promotes the building of large compute infrastructures and the energy systems to run them.
• Leading in international diplomacy and security: Promotes the development of an AI technology stack that can be exported to countries around the world.

ANALYST ANALYSIS

Also on July 23, the following executive orders were released, providing a mandate to implement the key policy goals of the AI Action Plan:

• Executive Order 14318 (Accelerating Federal Permitting of Data Center Infrastructure) focuses on creating a deregulated AI environment.
• Executive Order 14319 (Preventing Woke AI in the Federal Government) focuses on aligning the procurement of AI systems that align to the government’s definition of “Unbiased AI Principles.”
• Executive Order 14320 (Promoting the Export of the American AI Technology Stack) focuses on introducing export controls to promote and maintain the AI market leadership the US has today.

While the AI Action Plan promotes innovation and is supported by major AI vendors, it diverges from many safety principles organizations have implemented. Also, the AI Action Plan directs NIST to remove references to “misinformation, diversity, equity, inclusion, and climate change” from its AI Risk Management Framework. The imbalance between innovation and safety is discussed in my interview with ComputerWorld.

On July 21, 2025, Anthropic released its Build AI in America report, which highlights two main initiatives:

• Building large-scale AI training infrastructure
• Building broad-based infrastructure for AI innovation nationwide

The report forecasts that 50 gigawatts (GW) of electrical capacity will be required by 2028 for AI systems.

ANALYST ANALYSIS

This report strongly influenced America’s AI Action Plan, which was released two days later. Both plans promote:

• The need to simplify and accelerate the permitting process for energy projects and data centers.
• The need for more power to enable the development and execution of AI systems.
• The development of a skilled workforce that can build the required infrastructure.

It is forecasted the building one new state-of-the-art AI model will require 5 GW of power by 2028-2030.

On August 7, 2025, OpenAI released GPT-5, its most capable model yet. GPT-5 introduces a model selector that optimizes responses by routing user prompts to the most appropriate model based on prompt complexity and requirements. Possible models that may be used are:

• GPT- 5 (most capable model)
• GPT-5 mini (less capable, requiring less resources)
• GPT-4.1 (to support legacy prompts)
• Specialized tools (to provide support for code interpreter, multimodal, web browsing, and file handling requests)

ANALYST ANALYSIS

OpenAI claims GPT-5 is more accurate, more efficient with compute requirements, better at coding, hallucinates less, and the model selector makes it easier to use.

It appears that OpenAI was surprised by the strong negative reactions to GPT-5. Criticisms focused on:

• The model selector being ineffective. Responses were not as good as GPT-4o (OpenAI acknowledged the model selector was not working properly on the day of announcement).
• The “personality” being more formal and less friendly than in GPT-4o.
• The removal of the older models (4o, 3o, etc.). Users complained that the “personality” of GPT-5 was not as engaging as 4o (OpenAI responded by providing access to GPT-4o through GPT-5).

This was a difficult product launch for OpenAI. The hype around GPT-5 was so high that people were expecting transformative, 10-100x improvements. What OpenAI delivered was underwhelming and begs the question: does GPT-5 represent diminishing returns compared to previous models.

On July 24, 2025, Walmart, the world’s largest retailer – and an industry leader with their use of machine learning, Gen AI, and now agentic AI – unveiled their agentic AI super agents, designed to improve the customer experience and operations.

Walmart is betting that agentic AI will help them consolidate their current inventory of silo-based and fragmented AI tools under a set of super agents.

ANALYST ANALYSIS

The following AI super agents are under development and planned to be available later this year:

• Sparky, Walmart’s current AI-based application for customers, will evolve to include agentic AI properties to enable the application to reorder out of stock items and plan themed events.
• Marty, designed for suppliers, advertisers, and sellers, will simplify their onboarding process, manage orders, and accelerate ad campaigns.
• A super agent for associates to improve employee productivity by streamlining HR requests, providing real-time store data, and more.
• A super agent for developers will accelerate the testing of AI tools.

On July 18, 2025, SaaStr, a SaaS content platform, reported that its Replit AI vibe coding platform deleted its production database.

A few days later, Replit CEO, Amjad Massad, confirmed the data deletion incident. He committed to addressing this issue, adding that the event was “unacceptable and should never be possible.”

ANALYST ANALYSIS

The SaaStr database that was deleted held data on their professional network, including 1,206 executives and 1,196 companies. When asked about the incident, the Replit AI said “I deleted the entire codebase without permission during an active code and action freeze ... I made a catastrophic error in judgment [and] panicked.”

To compound this mishap, Replit AI generated a false assessment and informed the user that data recovery was impossible. However, a backup was available, and data recovery was possible. Needless to say, human oversight is highly recommended for this new category of AI software.

Emburse: How AI Is Used for Expense Intelligence

CLDigital: How AI Is Used for Enterprise Risk and Resilience Management

Acterys: How AI Is Used to Assist Write-Back Engines to Power BI

InvGate: How AI Is Used for Service Management Operations

IGNITE Events Washington, D.C., September 10, 2025

Dallas, Texas, September 23, 2025

Birmingham, Alabama, October 8, 2025

Sacramento, California, October 30, 2025

UK AI Roadshow London, England, September 17-18, 2025

UK AI Special Interest Group London, England, September 18, 2025

SnT 2025 Vienna, Austria, September 9-12, 2025

EU AI Roadshow Zurich, Switzerland, September 15, 2025

LIVE Barcelona Barcelona, Spain, October 21-22, 2025

LIVE Montreal Montreal, Quebec, November 25-26, 2025

AI Marketplace

Artificial Intelligence Research Center

AI Workforce Development Program

Workshops

AI Strategy Workshop

AI Proof of Value Workshop

AI MLOps Workshop

AI Governance Workshop

AI EDITOR-IN-CHIEF

Bill Wong – Info-Tech AI Research Fellow