How I Used Two AI Agents to Plan My Kids' Financial Future

Most people use AI to write emails or summarize articles. Last week, I used it to solve a €10,000+ investment decision that will impact my three children over the next 17 years.

The question was simple:

• Should I invest in a traditional PPR (Portuguese retirement plan) or a global ETF for my kids?

The answer was anything but simple. Between custody fees, management costs, compound interest calculations, and Portuguese tax law, I was looking at several variables across three different timelines.

I didn't just ask one AI to solve it. I orchestrated two different AI agents to get the best of two worlds:

• Gemini handled the research and fact-checking.
• Claude built the financial models.

And I brought the critical insight that neither AI would have found on its own.

This is the short story of how AI orchestration works in practice, and why the most valuable skill isn't prompting AI, it's managing a team of them.

Disclaimer: This article does not provide financial advice. It documents my personal research process for educational purposes. Consult qualified financial and tax professionals before making investment decisions.

The Problem: Three Kids, Three Timelines, One Decision

I have three children:

• Child 1: 6 years old (12 years until age 18)
• Child 2: 4 years old (14 years until age 18)
• Child 3: 1 year old (17 years until age 18)

I wanted to invest €100 per month for each of them, compounding until they reach adulthood. The traditional advice pointed to two options:

Option A: PPR (Plano Poupança Reforma)

• Provider: Casa de Investimento, Save and Grow
• Low taxation at exit (8.6%)
• Management fees (1.4% + 0.16% = 1.56% annually)
• Simple, widely recommended

Option B: Global ETF (Vanguard All World)

• Low annual costs (0.19% TER)
• Taxation at exit (19.6% on capital gains after holding period)
• More complex, requires custody account

On the surface, the PPR looked better because of the tax advantage. But something felt off about paying 1.56% annual fees for 17 years.

Every comparison I found online made different assumptions:

• What custody fees would I actually pay at Banco BiG vs ActivoBank in 2026?
• How much does the €12 trading commission at BiG matter if I only trade quarterly?
• What if interest rates drop from 8% to 5%?

I needed a way to model all these variables dynamically.

The Human Insight: The Donation Strategy

Before bringing AI into this, I had one critical piece of information that changed everything.

I'd been listening to the Contas-Poupança podcast, where they discussed a little-known strategy: donating ETFs to your children.

Here's how it works:

When you donate ETF shares to someone, the acquisition cost resets to the current market value. This means:

• You invest €100/month for 17 years
• The portfolio grows to €35,000
• You donate the shares to your child at age 18
• Your child receives the full €14,000 in capital gains and pays zero taxes

This wasn't something Gemini or Claude suggested. This was human research that I brought to the table. The AI's job was to help me calculate whether this strategy, combined with lower ETF fees, would beat the PPR—even with the hassle of custody accounts and trading fees.

This is what separates useful AI workflows from AI theater: expert outputs require expert inputs. I didn't ask AI to "figure out the best investment strategy." I came with a hypothesis and asked AI to help me stress-test it with real numbers.

Phase 1: Gemini as the Research Investigator

I started with Gemini for one reason: it has real-time web access and integrates directly with Google Search.

My brief was specific:

"I need you to verify 2026 banking fees for Portuguese brokers. Specifically:

• ActivoBank: custody fees, trading commissions, internal transfer fees
• Banco BiG: custody fees, trading commissions
• PPR management fees at major providers
• Current ETF taxation rules in Portugal

Cite every source. I need to trust these numbers."

Gemini came back with detailed breakdowns:

ActivoBank (2026 pricing):

• Custody: €44,28 annually
• Trading commission: €0 for 1 trade/month

Banco BiG:

• Custody: ~€25 annually
• Trading commission: €12 per trade

PPR:

• Management fees: 1.56% annually (1.4% + 0.16% at Casa de Investimento)
• Exit taxation: 8.6% on gains

ETF:

• TER (fund operating cost): 0.2% annually (Vanguard All World)
• Taxation: 19.6% on capital gains
• Donation strategy: 0% tax if executed properly

What made this valuable wasn't just the numbers. It was the sourcing. Gemini linked to official preçários (price lists) from each bank, dated 2026. This gave me confidence I wasn't working with outdated information or hallucinated fees.

The key finding: ActivoBank's zero-custody model meant I could invest monthly at zero cost, eliminating one of the major ETF disadvantages.

Phase 2: Claude as the Financial Architect

Once I had verified fees, I needed to hand off to Claude. But here's a crucial detail: I didn't just copy-paste my conversation with Gemini.

Instead, I asked Gemini to create a detailed summary of everything we'd discovered—all the fees, assumptions, and requirements—formatted specifically as a brief for Claude. This is AI orchestration in practice: one agent prepares the perfect input for the next agent.

Here's why I moved to Claude:

Gemini is great at research, but Claude excels at complex logic and document creation. I needed a dynamic financial model that would:

• Calculate compound interest monthly over 17 years
• Track three separate portfolios (different timelines for each child)
• Compare PPR vs three different broker options
• Update automatically when I changed assumptions

My brief to Claude was structured:

"Create a Google Spreadsheet comparing PPR vs ETF for 3 children with these specifications:

Variables (all editable):

• Monthly investment per child: €100
• Annual return: 8%
• PPR: 1.56% management fee, 8.6% exit tax
• ETF: 0.2% TER, 0.15% donation cost
• Three brokers to compare (BiG, ActivoBank, Other)

Calculations:

• Monthly compound interest over 12, 14, and 17 years
• Broker fees applied at correct intervals (BiG: quarterly trades, ActivoBank: monthly)
• Final comparison showing net value after taxes/fees"

Claude delivered a multi-sheet Excel file with:

• "Variáveis" sheet: Single source of truth for all assumptions (in blue, following financial modeling conventions)
• Three portfolio sheets: Month-by-month projections for each child
• Comparison sheet: Final net values side-by-side

The magic was in the formulas. Claude built proper compound interest calculations:

Month 1: =Investment × (1 + ((1 + Return - Fees)^(1/12) - 1))
Month 2: =PreviousBalance × (1 + MonthlyReturn) + Investment

This wasn't just a static calculation. When I changed the annual return from 8% to 5%, all 204 months recalculated instantly.

The Results: Numbers Don't Lie

Here's what the model revealed for my youngest child (Child 3, 17-year horizon):

Difference: €4,171 in favor of ETF

Scale this across all three children, and the advantage grows to over €7,000.

The ActivoBank advantage came from two factors:

1. Monthly investing at zero cost (vs BiG's quarterly trades with €12 fees)
2. Allow to invest monthly improving the DCA (Dollar cost averaging)

But here's what made the spreadsheet valuable: I could test scenarios.

• What if returns drop to 5%? ETF still wins by €3,200.
• What if I want the security of Banco BiG? ETF wins by €6,500 (3 kids).
• What if I increase monthly investment to €250? The gap widens to €27,000 (3 kids).

The spreadsheet:

Phase 3: Closing the Loop with Gemini

The final step was verification. I took Claude's mathematical logic back to Gemini:

"Review these compound interest formulas. Are they calculating monthly compounding correctly? Check the tax treatment of the donation strategy."

This wasn't just paranoia. It's good practice. AI models can be confident and wrong. Having a second AI peer-review the first one's work catches errors before you commit real money.

Gemini confirmed:

• Compound calculations were correct
• Tax treatment matched Portuguese law
• Donation timing assumption was accurate (with the 2-year lookback caveat)

This is the audit phase that most people skip. It's tempting to take the first AI's output as gospel. But in financial decisions, verification isn't optional.

What This Reveals About AI Orchestration

This wasn't about "AI solving my problem." This was about managing a team of specialized agents.

Here's the workflow pattern:

1. Human brings domain insight

I discovered the donation strategy through independent research. Neither AI would have suggested this on its own.

2. Gemini: Research & fact-checking

Used for real-time data gathering, fee verification, and source citation. Gemini's strength is current information.

3. Claude: Technical execution

Used for complex document creation, formula logic, and structured modeling. Claude's strength is architectural thinking.

4. Gemini: Verification loop

Back to Gemini for peer review of Claude's technical work. Catches errors and confirms assumptions.

5. Human makes final decision

The spreadsheet doesn't make the choice. It eliminates uncertainty so I can decide with confidence.

This pattern applies beyond financial planning:

• Market research → Data analysis → Report writing
• Legal research → Contract drafting → Compliance review
• Technical investigation → Code generation → Security audit

The future of AI work isn't about finding "the best AI." It's about orchestrating multiple agents for their specific strengths.

The Takeaway: From Prompts to Orchestration

Most AI tutorials teach you how to write better prompts. That's the wrong frame.

The valuable skill is workflow design: knowing when to use which tool, how to hand off between agents, and where human judgment is non-negotiable.

In this case:

• I could have verified 2026 banking fees manually, but it would have taken hours
• I could have built a dynamic compound interest model in Excel, but it would have taken days
• But AI couldn't have known about the donation strategy (requires domain research)

The human role isn't to do everything or to let AI do everything. It's to be the conductor.

This is what AI orchestration looks like in 2026. Not a single magic prompt. Not an autonomous agent that "does it all." A deliberately designed workflow where each participant—human and AI—contributes what they do best.

Beyond Financial Planning: The Pattern

The orchestration pattern I used applies to many complex decision-making scenarios:

Investment Portfolio Analysis

Research market conditions with one AI, model portfolio scenarios with another, verify tax implications with a third. Perfect for comparing retirement strategies, real estate investments, or business valuations.

Business Strategy Planning

Gather competitive intelligence, build financial projections, validate assumptions. One agent researches market data, another builds scenario models, a third reviews for logical flaws.

Technical Architecture Decisions

Research available technologies, model cost implications across vendors, verify security compliance. Essential for migrating systems, choosing platforms, or evaluating build vs. buy decisions.

The Common Pattern

What makes orchestration work across these domains:

1. Specialized agents for specialized tasks - Don't ask one AI to do everything
2. Structured handoffs - One agent prepares perfect input for the next
3. Verification loops - Cross-check critical calculations and assumptions
4. Human domain expertise - You bring insights AI can't discover alone
5. Dynamic modeling - Build tools that let you test "what if" scenarios

The value isn't just in the final answer—it's in building reusable frameworks for complex decisions.

Conclusion: The Conductor's Advantage

The €10,000 question wasn't really about PPR vs ETF. It was about learning to orchestrate intelligence—human and artificial—to solve problems that matter.

Most people will use AI as a better search engine. Some will use it as a writing assistant. But the real leverage comes from treating AI as a team of specialists, each excellent at their specific role, coordinated by someone who understands both the domain and the workflow.

The technology exists today. The patterns are proven. What's missing is the mindset shift: from "prompting AI" to "conducting AI."

Your Next Step

Pick a complex decision you've been avoiding because the analysis seems overwhelming. It could be:

• Comparing insurance policies or investment options
• Evaluating vendors for a business purchase
• Planning a major life financial decision

Break it into phases: research, modeling, verification. Assign each phase to the right tool. Be the conductor.

Start this week. The decisions you've been postponing might be simpler than you think.

Important Note on Donation Timing:

The tax reset from donating ETFs has a 2-year lookback rule. If you donate when your child is 20 years old, and you started investing when they were 1, you're exempt from taxes on all gains from age 1 to 18. However, gains accumulated in the last 2 years (age 18-20) remain taxable. For maximum benefit, donate at exactly age 18 to avoid any taxable period.

Disclaimer:

This article does not provide financial advice in any form. It describes my personal research process for educational purposes only. Tax laws change frequently, and individual circumstances vary significantly. The donation strategy has specific timing requirements (2-year lookback rule as noted above) and may not be suitable for your situation. You must consult with qualified financial advisors and tax professionals before making any investment decisions. Past performance does not guarantee future results.