Recent Summaries

This newsletter, sponsored by Rimini Street, discusses the evolution of ERP (Enterprise Resource Planning) systems and how the rise of composable architectures and agentic AI are revolutionizing them. The key argument is that businesses are moving away from monolithic ERP systems towards more flexible and interoperable solutions that prioritize business needs over vendor lock-in, leading to significant performance gains.

• Evolution of ERP: From mainframes to cloud, ERP systems have constantly adapted to technological advancements.

• Composable Architectures: Enable organizations to select and combine capabilities from various systems, breaking free from vendor dependence.

• Agentic AI: Facilitates coordination between systems that were not originally designed to communicate, automating cross-platform operations.

• Performance Gains: AI-enabled ERP systems can significantly boost user satisfaction, productivity, processing time, and decision accuracy.

• Agentic AI acts as a UX and orchestration layer, streamlining workflows across different systems.

• The shift allows businesses to organize technology around their needs, rather than adapting their processes to the ERP system.

• Early adopters of AI-driven ERP systems stand to gain a significant competitive advantage.

• Enterprises can modernize their existing systems by reconfiguring and extending them, instead of relying on complete ERP upgrades.

This Gradient Flow newsletter focuses on the critical challenges of deploying AI systems into production, highlighting the gap between successful pilot projects and reliable, real-world applications. It uses medical AI as a case study, illustrating common failure modes and offering a playbook of solutions applicable across industries. The core argument is that reliability, not just raw performance, is paramount for successful AI deployment, and that this requires a shift in focus towards system design and human-AI collaboration.

• AI Production Gap: Many AI projects fail to move beyond the pilot stage due to reliability issues.

• Reliability over Performance: Predictability and consistency are more important than peak accuracy in real-world AI applications.

• Medical AI as a Model: The high-stakes nature of medical AI provides valuable lessons for building reliable AI systems in other domains.

• Human-AI Collaboration: AI should augment, not replace, human expertise, with structured workflows and appropriate oversight.

• Hallucinations and Inconsistency: Generative AI models can produce false information and inconsistent recommendations, posing significant risks.

• Automation Bias: Over-reliance on AI outputs can lead to errors, as demonstrated in a mammography trial where radiologists missed malignancies when using AI assistance.

• Knowledge Grounding and RAG: Techniques like Retrieval-Augmented Generation (RAG) can improve reliability by grounding AI responses in vetted sources.

• Uncertainty Management: AI systems should be designed to know their limits and abstain from answering when uncertain, routing queries to human experts.

• Deliberate System Design: Transitioning from pilot to production requires prioritizing predictable behavior and carefully structuring human oversight.

OpenAI is making significant changes to ChatGPT's accessibility and monetization strategy. This includes introducing a cheaper subscription tier with ads while simultaneously assuring premium users of ad-free experiences. These moves come as the company faces scrutiny over substantial financial losses.

• Tiered Access: ChatGPT is now offered in tiers: Free (with ads), Go ($8/month with ads), Plus ($20/month, ad-free), and Pro ($200/month, ad-free), each catering to different user needs and usage intensity.
• Monetization Shift: OpenAI is embracing advertising in free and lower-tier subscriptions, a notable change in strategy for revenue generation.
• Ad Policy: OpenAI pledges to keep advertising separate from AI responses, ensure data privacy, and offer options for ad personalization control and ad-free experiences for premium users.
• Global Expansion: The "ChatGPT Go" plan, initially launched in India, is now available in over 170 countries to improve access to AI, with pricing adjusted for local markets.
• Strategic U-Turn: The introduction of advertising represents a shift from CEO Sam Altman's previous stance against integrating AI and advertising.

This newsletter covers a range of tech-related topics, from the US banning individuals fighting online hate to the rise of AI companions and the creation of new emotions via AI. It also touches on the potential bursting of the AI bubble and the challenges of mining Greenland's natural resources, offering a snapshot of the current state and future trends in technology and its societal impact.

• Digital Rights Crackdown: The US government is escalating its stance against digital rights, banning individuals involved in combating online hate, raising concerns about censorship and politicization of online safety.

• AI Companionship: AI chatbots are increasingly used for companionship, especially among teenagers, but concerns arise regarding their potential to exacerbate mental health issues and the need for regulation.

• Emergence of Neo-Emotions: People are using AI to invent and define new emotions, reflecting a growing need to articulate nuanced aspects of feeling in the digital age.

• AI Bubble Concerns: The newsletter questions the longevity of the AI boom, discussing the likelihood of a bubble burst and what valuable applications will remain.

• Geopolitical Tech Challenges: The newsletter highlights challenges such as Iran's internet shutdown and the difficulties in mining Greenland's resources, showing the intersection of technology with geopolitics and environmental constraints.

• The banning of individuals fighting online hate suggests a concerning shift in the US government's approach to digital rights and online content moderation.

• The rise of AI companions presents both opportunities and risks, emphasizing the importance of ethical considerations and regulation in the development of AI technologies.

• The creation of new emotions via AI highlights the evolving relationship between humans and technology and the potential for AI to shape our understanding of emotions.

• The potential bursting of the AI bubble serves as a reminder of the cyclical nature of technology hype and the need for realistic assessments of AI's capabilities.

• The newsletter's coverage of diverse topics like internet shutdowns, resource mining, and romance scams highlights the broad impact of technology on society, politics, and the environment.

This Latent Space newsletter features an interview with Brex CTO James Reggio, detailing the company's AI-driven turnaround from stalled growth in 2024 to exceeding $500M ARR in 2025. It highlights Brex's cultural shift towards AI fluency, its modernized tech stack, and the application of AI agents to streamline operations.

• AI-First Culture: Brex fostered a company-wide AI adoption through training programs, fluency levels, and incentivized use of AI tools, even revamping their hiring process to evaluate AI competence.
• Modernized Tech Stack: Brex embraced modern tools and frameworks, including Typescript, Mastra, pgvector, Pinecone, Greptile, and Sierra, to build their AI agent platform. They also provide employees with a choice of coding tools and LLMs through ConductorOne.
• Agent-Based Architecture: Brex moved from a monolithic agent to a multi-agent system with an orchestrator, enabling specialized sub-agents to handle specific tasks and collaborate for complex workflows.
• Strategic AI Implementation: Brex structured its AI strategy around three pillars: Corporate AI (internal tooling), Operational AI (cost reduction), and Product AI (customer-facing features), with a focus on automating processes while maintaining high customer satisfaction.
• Importance of Experimentation: Brex emphasizes the value of trying multiple approaches, even those that fail, to learn and discover optimal solutions in the rapidly evolving AI landscape, citing their unsuccessful attempt with reinforcement learning for credit decisions as a key learning experience.

This newsletter from MIT Technology Review covers the debate around AI-powered coding, breakthrough biotech trends, and the future of AI. It also touches on topics like misinformation, online child abuse, and the global humanoid robot market.

• AI Coding Hype vs. Reality: Explores the mixed opinions on AI coding, with some seeing it as a productivity booster and others as a source of poorly designed code.

• Biotech Breakthroughs: Highlights gene editing, ancient gene resurrection, and controversial embryo screening as key biotech trends.

• AI's Future: Discusses five hot trends to watch in AI, including the potential and pitfalls.

• Ethical Concerns with AI: Covers the troubling use of AI in generating child abuse material and instances of chatbots providing harmful advice.

• The impact of AI on coding is still uncertain, with developers and executives holding differing views on its benefits and drawbacks.

• Biotech advancements raise ethical questions, particularly regarding embryo screening and genetic manipulation.

• Misinformation and content creation are increasingly merging, posing challenges for governance and public discourse.

• China is currently dominating the global humanoid robot market, accounting for a significant portion of deployments.