The Problem With AI Naming Conventions
Dr. Brian Massey
Helping FS Leaders Solve The Organisational Challenges Of AI | Managing Partner at Anordea | Strategy Advisor, Professor & Keynote Speaker
Hello and welcome to the latest edition of Banking On AI Strategy. In this edition we’ll review some recent industry developments in AI, and consider the problems that ambiguous AI naming conventions are creating for users.
AI Industry Updates:
Policy: China Seeks Allies in AI Race
China has invited the United Kingdom to participate in the upcoming World AI Conference in Shanghai this July, signaling a desire for closer collaboration on global AI governance. The invitation was extended by Chinese Ambassador Zheng Zeguang during a speech in London, where he highlighted recent joint activities such as the UK’s participation in a China-hosted AI workshop and China’s involvement in the 2023 AI Safety Summit in the UK. While advocating for deeper cooperation on emerging technologies like AI, Zheng criticized what he called “ideological biases” and national security concerns in UK-China relations, and warned against protectionist policies, including recent trade deals with the U.S. that could affect Chinese firms’ access to British supply chains.
This invitation highlights the geopolitical complexity surrounding global AI governance, particularly for financial institutions operating across both Western and Eastern markets. As China seeks to shape the norms and standards of AI development, FS leaders will need to navigate divergent regulatory philosophies, data policies, and trust frameworks between blocs. The UK’s engagement with both U.S. and Chinese AI ecosystems places its financial services sector in a delicate position—potentially benefiting from broader innovation access, while also facing compliance, security, and ethical scrutiny from competing global powers. Institutions must monitor how international alignment (or fragmentation) around AI governance could influence cross-border data flows, vendor partnerships, and the deployment of AI-driven services in regulated markets.
Technology: AI That Codes Itself Is Here
Google’s new AI doesn’t just write code, it discovers new algorithms to make code run faster. This week, DeepMind introduced AlphaEvolve, a self-improving agent that’s already saved millions in costs by optimizing Google’s internal systems. And it is AI that codes itself. Powered by Gemini models, AlphaEvolve combines the creative reasoning of large language models with automated evaluators to evolve complex codebases and propose high-performance, verifiable solutions across mathematics, chip design, data center scheduling, and AI training. It has already delivered real-world gains, recovering 0.7% of Google’s compute resources, improving TPU design, and accelerating AI kernel performance by up to 32.5%. Beyond infrastructure, AlphaEvolve has contributed to open mathematical problems, even advancing the centuries-old “kissing number” challenge in higher dimensions.
AlphaEvolve exemplifies the emerging wave of agentic AI, systems that not only generate ideas but rigorously test, evolve, and deploy them autonomously. For financial services leaders, this signals a powerful new toolset: imagine agents that can optimize trading algorithms, evolve compliance rule engines, or streamline operational logic in real time. The integration of generative creativity with automated verification will be crucial for building trustworthy AI in highly regulated environments. As FS firms explore AI co-pilots and agentic workflows, AlphaEvolve-style systems offer a glimpse of what’s next: AI not as a static tool, but as a strategic partner capable of driving continuous optimisation, innovation, and transformation across core business functions.
Competition: The Battle Of The AI Agents Begins
OpenAI has unveiled Codex, a new AI agent tailored for software engineering tasks such as writing features, fixing bugs, and running tests. Initially available as a “research preview” for paid ChatGPT users, Codex represents OpenAI’s push into the competitive market of coding assistants already populated by Google, Microsoft, Anthropic, and startups like Anysphere and Windsurf. Notably, OpenAI is in acquisition talks to purchase Windsurf for around $3 billion, underscoring the strategic value it places on AI tools that support developers. Codex, powered by a variant of the o3 model optimized for software reasoning, is already in daily use by OpenAI engineers and partners like Cisco and Kodiak Robotics, signalling rapid progress toward OpenAI’s broader agentic AI ambitions.
Codex reflects a growing shift in AI capabilities from static assistance to dynamic, task-completing agents—raising key implications for the financial services sector. Banks and asset managers increasingly rely on software pipelines for everything from regulatory automation to portfolio modeling, and tools like Codex could help rapidly accelerate these workflows while reducing operational costs. However, the rise of autonomous coding agents also introduces governance and security challenges—especially in tightly regulated environments. Leaders in FS must prepare for a future where internal systems, compliance engines, and trading platforms are increasingly co-developed or maintained by AI, making oversight, explainability, and responsible AI policies more critical than ever.
The Problem with AI Naming Conventions
An increasingly common source of frustration in the AIU space is the confusing and inconsistent naming of AI tools. As the number of available models and services grows, it’s becoming harder to identify what a tool actually does, or whether it’s the right fit, based on its name alone. This challenge is more than just an annoyance; it has real implications for leaders trying to drive AI transformations effectively. In this article, we’ll unpack why naming conventions are becoming a problem and explore what leaders should consider as they navigate this evolving landscape.
Why Naming Matters
Names matter. From the earliest writings on myth and magic, knowing the name of someone or something is supposed to give you some power over it. There is some deep psychological truth to this and we know from communication theory that using someone’s name when speaking to them is a fast way to gain their attention and focus. So, we can establish that names are important. When it comes to working with AI however, names present us with a real problem in that they do not give us any insight into the tools we are working with. This is one area where the anthropomorphisation of AI has done us a big disservice. We need something more than just a simple placeholder and ideally names should communicate some aspect of the particular AI tool we are using and how it will help us achieve our work goals and aims. However, currently the opposite is true. AI naming conventions are irrational, dispersed, and appear to follow no singular logic. To understand this better let’s look at some of the most popular AI tools available today and not just their names, but also how they are named relative to one another.
The Current Landscape of AI Naming
Looking at the evolution of AI naming conventions over recent years, we can see some common themes. People are beginning to get broadly familiar with the terms copilot, assistant, agent, bot, and even GPT. Across different firms that create foundation models or implement them into service offerings such as Open AI, Microsoft, and Google, these terms crop up again and again. In fact, the term copilot is probably the most ubiquitous term in AI at the moment, where every software package offers some version of a copilot, each fulfilling different roles and functions. The complexity appears to be growing and naming conventions amongst the offerings of companies are growing more confusing too. As of writing the current offerings from OpenAI include GPT-4o, o3-Mini, o4-Mini, o4-Mini-High, GPT-4.1, GPT-4.1-Mini, GPT-4.5, in just the basic chat interface. It’s starting to look like we need an AI tool just to figure out which AI tools we should be using day to day.
To continue reading this article, and to receive all future editions direct to your inbox, subscribe for premium or free access on Substack today at:
Thank you for this, Brian. Interesting read as always.