- The Shift Brief
- Posts
- GPT-5.2 Is Great. That's Exactly the Problem.
GPT-5.2 Is Great. That's Exactly the Problem.
The Shift Brief | Week of Dec 15th, 2025
Ryan Erickson
December 16, 2025
Why Your AI Strategy Needs to Be Model-Agnostic
If it feels like there's a new AI announcement every week, you're not imagining it.
New models. New benchmarks. New claims about what AI can now replace or automate. For leadership teams inside financial institutions, this creates real tension. You don't want to pick the wrong vendor or lock into the wrong approach. But you also can't afford to stand still.
The mistake many firms are making is treating the model layer as the strategy.
The safer bet is the opposite.
The Problem With Waiting for Someone Else to "Figure It Out"
A common mindset we hear is: "We're a Microsoft shop. We'll wait for Copilot to mature."
That sounds reasonable on the surface. In practice, it's become a form of strategic paralysis.
Microsoft recently signaled that it's pulling back on some Copilot investment as adoption has lagged expectations. That doesn't mean Copilot is going away. It does mean that broad, horizontal AI tools dropped into existing software aren't magically solving real workflow problems.
If your firm has been waiting for Microsoft, or any single vendor, to fully "figure it out" on your behalf, you may be waiting a while.
AI adoption is proving to be far more contextual. It depends on data quality, workflow design, and how well tools integrate with how teams actually work. No platform vendor can solve that generically.
GPT-5.2 Is Impressive. That's the Point.
This week's GPT-5.2 release is a good example of why the ground keeps shifting.
Dave Wang recently published a strong breakdown of what actually matters in the release, especially for finance teams. Instead of hype, he focused on where the improvements are genuinely usable.
A few highlights from his analysis:
Meaningful accuracy gains on core finance tasks like DCFs and three-statement models
Much larger context windows, enough to work across multiple filings in a single workflow
Lower hallucination rates in financial and tax domains when browsing is enabled
In his testing, GPT-5.2 could generate a solid first-pass DCF model in minutes rather than hours, getting analysts most of the way there before human judgment and stress testing take over.
Dave's newsletter is excellent and well worth following: 👉 https://www.davewang.ai/
Anthropic Is Sending the Same Signal
Anthropic's recent announcements point in a similar direction.
Their focus on open foundations and compatibility with MCP isn't about winning a single benchmark cycle. It's about building systems that can work across models, tools, and future architectures that don't yet exist.
That's a strong signal. The builders closest to the frontier are planning for a world where no single model, or vendor, stays dominant for long.
The Strategic Takeaway
The firms that win in this environment won't be the ones who picked the "best" model in 2025.
They'll be the ones who built foundations that let them change their minds.
Clean data. Strong governance. Flexible pipelines. Clear separation between data, logic, and models. Those are the things that make an organization AI-agnostic.
Once you have that, model releases become an advantage instead of a disruption. You can test GPT-5.2 today, something else tomorrow, and whatever comes next without re-architecting your entire stack.
Final Thought
AI is moving too fast to wait for a single vendor to solve it for you.
Model leadership is temporary. Platform roadmaps slip. Adoption is uneven.
Optionality is the strategy.
~Ryan Erickson, Founding Executive