The News
Last week, Google announced a set of upgrades to their Gemini API that the developer world is calling a game-changer. The big one: "tool combos" — the ability for AI to combine multiple tools, APIs, and data sources in a single request instead of making separate calls for each step.
Google Search, Google Maps, your own custom backend — all callable in one pass. One prompt. One response. Everything your AI needs to answer a complex question, executed in a single shot.
They also shipped "context circulation" — carrying reasoning context across tool calls and conversation turns so the AI doesn't lose track of what it's doing mid-workflow.
The tech press is impressed. Developers are excited. And if you're running a law firm, you should pay attention — but not for the reason you think.
We've Been Running This Since October
What Google just shipped to developers as a new feature, we deployed to law firms in Q4 2025.
We call it Programmatic Tool Calling. The concept is the same: instead of making your AI execute one step at a time — pull a case file, wait, check the ledger, wait, draft an email, wait, update the CRM, wait — the system executes the entire workflow in a single inference pass.
Sequential (Everyone Else)
- Step 1: Look up client → wait
- Step 2: Pull case file → wait
- Step 3: Check payment status → wait
- Step 4: Draft follow-up → wait
- Step 5: Update CRM → wait
- 19+ separate AI calls per workflow
- Slower. More expensive. More failure points.
Single-Pass (NB OS)
- One prompt triggers the full workflow
- Client lookup + case file + payment + email + CRM
- All executed in a single inference pass
- Context preserved across every step
- Result delivered as one complete output
- 1 AI call does the work of 19
- Faster. Cheaper. More reliable.
This is not a theoretical architecture. It runs in production, every day, processing real client intakes, real case follow-ups, and real collections workflows for real law firms.
Why This Matters for Your Firm
You don't care about inference passes. You care about two things: does it work, and what does it cost.
Single-pass execution means your AI intake process runs in seconds, not minutes. A new lead calls at 2 AM. The AI answers, qualifies them, books a consultation on your calendar, pulls their documents through GetDocs, and updates your CRM — before anyone on your team wakes up. One pass. Done.
It also means lower operating costs. Every separate AI call costs compute time and money. When you eliminate 19 calls and replace them with one, you cut your AI operating cost by more than a third. That's margin your competitors are burning.
The firms running NB OS today are already operating on the architecture Google wants developers to build toward tomorrow. You're not early — you're exactly on time. Everyone else is late.
What Google Got Right
Credit where it's due. Google's implementation is clean. Combining function calling with built-in search and maps in a single API request is smart. Context circulation across turns solves a real problem. And making it available on a free tier means more developers will build with this pattern.
That's good for us. More developers building multi-tool agents means more validation of the architecture. More people understanding why sequential calling is dead. More firms asking "why doesn't MY software work like this?"
The answer is: it can. It just requires someone who built it before Google made it fashionable.
What Happens Next
Every major AI company is moving toward single-pass multi-tool execution. Google shipped it this month. Anthropic (the company behind the AI that powers NB OS) has had it in their SDK since last year. OpenAI will follow. It's not a question of if — it's a question of who deploys it for YOUR industry first.
For law firms, we already did.
If your firm is still running on disconnected tools that don't talk to each other, where every AI interaction is a separate round-trip that costs you time and money — that's a problem with a known solution. And the world's largest technology company just confirmed it.