In mid-2025, OpenAI ran an experiment. They took dozens of knowledge-worker roles, broke them into task buckets, and had AI produce the same deliverables as subject-matter experts averaging 14 years of experience. A blind evaluation scored who did it better. The 2024 models tied or won 12% of the time. By mid-2025, that hit 47%. After the late-2025 model releases — ChatGPT 5.2, Claude Opus 4.5, Gemini 3 — it crossed 70%.
That’s not a trend line. That’s a threshold. And if you’re running a CAS practice with five to 50 people, you need to understand what just shifted — because the bottleneck in getting ROI from AI changed overnight, and most firms are still throwing money at last year’s problem.
The thing that was killing your AI efforts just went away
For the last two years, the experience of AI inside most practices has been the same. You sent staff to training. Some of them experimented. A few got excited. But nothing changed in the actual delivery work. Month-end still looked the same. The same manual processes, the same bottlenecks, the same rework.
That wasn’t a people problem. It was a quality problem. The AI output sat below what your experienced staff could produce — so everything it generated needed to be cleaned, checked, and often redone. The net effect was negative. Your team tried it, it made them slower, and they went back to how things were. Every firm owner I talk to has some version of this story.
That quality gap is gone. When the models produce work at or above your senior staff’s level — and do it 10 to 100 times faster — the rework disappears. The deliverable comes out finished. That’s the tipping point. Not a marginal improvement in a chatbot. A structural shift in what your practice can produce and how fast.
Your new problem is context — and it’s an operational one
Here’s the catch that matters for your firm. Those win rates only hold when the AI has all the context it needs — strategy, pricing, client history, SOPs, the way your team handles that one client who always sends source docs late. Feed it all of that, and the output is expert-grade. Feed it nothing — which is what most people do — and you’re back to the same generic output from 2024.
This is where it becomes your problem, not your team’s problem. The context that makes AI effective inside your practice — your workflows, your pricing logic, your service delivery standards, your advisory approach — most of it lives in people’s heads right now. It’s in scattered emails, half-finished process docs, and the institutional knowledge that walks out the door every time someone leaves.
If that context isn’t documented and structured, your AI tools can’t use it. And your competitors who do document it are about to operate at a speed and quality level you can’t match by adding headcount.
Artifacts build the moat
This is where the shift gets practical — and where it starts looking like enterprise value, not just productivity. Every deliverable your firm builds with AI — an SOP, a client dashboard, a workflow map, an interactive report — serves a dual purpose. It’s a finished product for your team and your clients, and it’s structured, machine-readable context that makes your AI smarter the next time it touches that workflow.
Before December 2025, the AI couldn’t produce artifacts that looked professional enough to deploy. That limitation is over. You can now build production-quality deliverables by describing what you want. The design bottleneck is gone. The development bottleneck is gone. What’s left is the hard operational question: can your practice articulate what it actually does clearly enough for a machine to act on it?
If you can, the math starts compounding. More documented context means better AI output. Better output means faster delivery. Faster delivery means more capacity for advisory work — the work where your real pricing power lives. And every artifact you produce along the way increases the transferable value of your firm.
You need two people, not one
One thing that’s become clear from working through this inside my own practice: you can’t hand AI implementation to a single person and expect results. You need two complementary mindsets working together.
You need someone who experiments relentlessly — tries new things, breaks them, moves fast, doesn’t care about failing. And you need someone who operationalizes what works — documents it, builds the repeatable process, makes sure the team actually adopts it. One without the other stalls. The experimenter generates breakthroughs but never systematizes them. The operator builds systems but resists the pace of change.
If you’re the owner, your job is to find this pair inside your firm and give them room to work. That’s the team that actually gets traction — not a firm-wide lunch-and-learn, not a Slack channel full of AI articles. Two people. One experiments, one operationalizes. That’s your AI implementation engine.
What this means for Q1
Stop investing your energy in the old constraint. Prompt engineering isn’t the bottleneck anymore. Training your team to write better prompts isn’t the highest-value activity you can fund this quarter. Context is.
Here’s what moves the needle right now. Start documenting your core workflows as structured artifacts — not in paragraph form buried in a shared drive, but as machine-readable context your AI tools can actually use. When something breaks, capture the conversation, generate the incident report, update the SOP. Turn the institutional knowledge inside your practice into something that compounds instead of something that evaporates.
You don’t need to overhaul everything by March. But you do need to stop solving a problem that no longer exists and start solving the one that does. Every week you wait is a week your competitors are building a context library you don’t have — and the gap between firms that have it and firms that don’t is about to get very wide, very fast.
The tipping point already happened. The question is whether your practice is positioned to operate on the other side of it.
That’s exactly what our AI Practice Transformation program is built for — a 15-day intensive that gets your Champion trained and your context built. No theory. A working system your team owns on day 16. Visit theaiaccountant.ai to apply.