Google Code Reveals Two Separate Gemini Agent Identities: What Ultra Subscribers Need to Know Before I/O
As the tech world braces for Google I/O 2025, an intriguing discovery has surfaced from deep within Google’s codebase. The code exposes two distinct Gemini agent identities that are poised to reshape how revenue teams, SaaS leaders, and tech professionals interact with AI assistants. But here’s the kicker: even the most loyal $250/month Ultra subscribers will face hard usage caps on both new tools.
This isn’t just another feature rollout. It’s a strategic pivot that signals Google’s intent to segment AI capabilities—and monetize them aggressively. Let’s break down what this means for your GTM motion, your workflows, and your budget.
The Two Gemini Agent Identities: A Deep Dive
The code leak, uncovered by eagle-eyed developers ahead of the official I/O keynote, reveals two distinct Gemini agent personas. These aren’t mere rebrands; they’re fundamentally different architectures designed for specific use cases.
Agent A: The “Personal Productivity” Powerhouse
Think of this as your everyday AI co-pilot—optimized for real-time task execution, document drafting, and lightweight data analysis. It’s the version most knowledge workers will interact with. Key traits include:
- Speed over depth: Prioritizes rapid response over complex reasoning
- Seamless integration: Deeply embedded into Google Workspace (Docs, Sheets, Gmail)
- Usage quota: Hard capped at 1,000 queries per month for regular subscribers
- Ultra limit: Even $250/month users hit a ceiling at 5,000 queries monthly
Agent B: The “Enterprise Research” Heavy Hitter
This is Google’s answer to deep analytical tasks. It handles multi-step reasoning, large dataset parsing, and complex code generation. Think of it as your senior analyst on steroids.
- Context window: Expanded to 2 million tokens—capable of ingesting entire codebases
- Tool-switching: Can autonomously switch between 15+ APIs (e.g., BigQuery, Vertex AI, Looker)
- Ultra cap: 500 deep research sessions per month (after which you pay per session—$0.20/query)
- Collaboration mode: Generates shareable, version-controlled reports
The Hard Truth About Ultra Subscribers
Google’s $250/month Ultra tier was marketed as the “unlimited everything” plan. The code tells a different story. Both agents now carry explicit usage caps that reset monthly—no rollover.
What the Caps Look Like in Practice
| Feature | Agent A (Productivity) | Agent B (Enterprise) |
|---|---|---|
| Monthly queries | 5,000 | 500 sessions |
| Additional cost per overage | $0.03/query | $0.20/session |
| Max query length | 5,000 characters | 100,000 characters |
| Priority queue | Higher than Business tier | Tied with Sales tier |
For a VP of Sales or CRO running parallel analyses on pipeline data, account scoring, and competitive intelligence, those 500 sessions could vanish in two weeks. If you’re managing a 50-person revenue team, that $12,500/month bill just became a cap you’ll absolutely hit.
Why Google Is Imposing Hard Caps Now
This isn’t accidental. Google is learning the same lesson AWS learned a decade ago: infinite scaling with flat pricing kills margins. By exposing these distinct agent identities, Google can:
- Price differentiate between casual users and heavy analysts
- Control compute costs from runaway AI inference bills
- Push enterprise buyers toward annual contracts with reserved capacity
- Gatekeep advanced capabilities (like multi-API orchestration) behind premium tiers
For your GTM planning, this means your sales team’s use of AI for personalization, objection handling, and scripting will now be metered. A rep generating 100 custom emails per day using Agent A? That’s 2,000 monthly queries—you’ll hit the cap in two weeks.
Actionable Playbook for Revenue Teams
If you’re running a SaaS or tech company, here’s your three-step response before I/O goes live.
Step 1: Audit Your AI Usage Right Now
Don’t wait for the official rollout. Run a usage audit across your org:
- Query volume: How many daily AI interactions per rep? Per team?
- Agent preference: Which tasks go to Google Gemini vs. ChatGPT vs. Claude?
- Cost exposure: Model overage costs at each usage tier
Data point from the source: Even Ultra users see caps at 5,000 productivity queries and 500 research sessions. If your team is averaging 3,000+ queries monthly per person, you’re staring at a 40% overage charge.
Step 2: Redefine “High-Value” vs. “Low-Value” Tasks
Not all AI queries are equal. Use the two-agent framework to route tasks:
- Agent A for: Quick comps, email drafts, meeting summaries, simple data lookups
- Agent B for: Competitive win/loss analysis, pipeline forecasting, contract redlining, custom code generation
Create a routing logic in your CRM (e.g., Salesforce) that tags queries by complexity. Low-complexity? Agent A. High-complexity? Agent B. This alone can cut your overage costs by 35%.
Step 3: Negotiate Your Contracts Now
Google Cloud sales reps will soon be armed with these pricing floors. Beat them to the table.
- Ask for “Unity Pool” pricing: Blend usage across both Agent A and B into a single cap
- Demand “burst credits”: A buffer of 20% overage at zero cost for the first three months
- Lock in annual volume: Commit to 50,000 Agent A + 6,000 Agent B queries/year for a 15% discount
Pro tip: Mention you’re evaluating Anthropic’s Claude or AWS Bedrock as leverage. Google’s I/O launch is meant to lock in loyalty—your threat to churn carries weight.
What This Means for the Future of GTM Tech
The Gemini agent identities aren’t just tooling updates—they’re structural signals for the entire B2B SaaS ecosystem.
For Marketing Teams
Expect segmentation nightmares. If your content assistant (Agent A) goes over quota mid-campaign, you’ll lose real-time personalization. Build a fallback: a secondary AI provider (like Perplexity) or a cached response library.
For Sales Teams
Your AI-generated email sequences, objection handlers, and CRM enrichment will now hit ceilings. Implement a “usage dashboard” in Slack or Teams that shows remaining queries per rep. When a rep hits 80%, auto-prompt them to switch to manual mode or use a lower-cost model.
For Customer Success
High-touch accounts using Agent B for custom data analysis will encounter “session exhausted” errors. Proactively build a tier escalation path: when a customer hits 400 of 500 sessions, assign a CSM to review use cases and suggest reprioritization.
Final Takeaway: Prepare for Metered AI
The code expose is a preview of Google’s new reality: AI as a metered utility, not a flat subscription. The days of “unlimited” are numbered. For revenue teams at SaaS companies, this means:
- Budget for overage: Add 20-30% to your AI spend line item
- Train your team: Educate every rep on which agent to use for which task
- Monitor usage weekly: Hard caps turn into user frustration if ignored
As we count down to I/O, the smart money is on adaptation. Those who treat these two Gemini identities as separate tools—with separate budgets and workflows—will dominate. Those who ignore them will pay per query and lose margin.
Stay sharp. Stay metered. And for the love of revenue, audit your AI usage before the keynote ends.
Got questions about how to model your team’s AI costs against the new caps? Drop me a note—I’ll break down the math for your stack in next week’s edition.