AI in finance: Vibe Excel reveals adoption inertia as new bottleneck
Financial modeling with ChatGPT and Claude reveals AI's growing pains in white-collar work
The history of AI in software engineering provides a clear blueprint. After ChatGPT's launch in late 2022, developers started copying code into the tool for generation and debugging. Anthropic's Sonnet 3.5 in mid-2024 marked a key inflection point, allowing AI to handle long-horizon tasks. By mid-2025, AI-assisted coding was widely adopted, with companies even engaging in 'tokenmaxxing' competitions. Coding was uniquely suited for AI due to verifiable outputs (tests) and existing tooling (IDEs, git).
Now, the author tests ChatGPT and Anthropic's Excel add-ons for typical investor use cases like revenue builds and scenario analysis. The experience feels similar to early AI coding: clearly useful but requiring heavy iteration. For example, building a financial model for an SPV into a pre-IPO AI company involved multiple rounds of manual tweaks. Key pain points include intent-output mismatch—short prompts fail to capture necessary context. The author concludes that while model capability was the main constraint for coding, for white-collar work, organizational trust and integration (adoption inertia) will be the bigger hurdle, offering a temporary edge for early adopters.
- AI coding went from novelty to ubiquity in ~2.5 years (ChatGPT Dec 2022 → mid-2025), driven by models like Sonnet 3.5 enabling long-horizon tasks.
- Finance is at a similar inflection point: ChatGPT and Anthropic's Excel add-ons show promise but require heavy iteration due to intent-output mismatch.
- For white-collar work, adoption inertia (organizational trust, integration) becomes the main bottleneck, not model intelligence.
Why It Matters
Organizations that overcome adoption inertia may gain a brief window of advantage in applying AI to non-obvious knowledge-work tasks.