Insight / Experiment

I Asked AI to Help Me Make 5 Difficult Decisions. Something Was Missing.

Can AI actually help us make better decisions — or is it mostly getting better at giving us better-sounding answers?

5 decisions Gemini + Quorum Deliberately challenged Decision process experiment

I wanted to answer a simple question:

Can AI actually help us make better decisions — or is it mostly getting better at giving us better-sounding answers?

So I took five difficult decisions that sit very close to real life:

  • ₹48L job vs a stable ₹34L job
  • Buying a ₹2.5Cr home vs continuing to invest
  • Moving abroad vs staying in India
  • Leaving a good career to build a company
  • How a family should balance career, money, ambition and family life

I ran each through Gemini and through Quorum.

I expected the systems to disagree.

They often didn't.

That's what made the experiment interesting.

The difference wasn't simply which AI produced the better answer. The difference was what happened after the answer. See what changed across all 5 decisions

The surprising finding

A good AI can already do something remarkable.

It can understand a complicated situation, identify trade-offs, challenge assumptions and give you a coherent recommendation.

But consequential decisions often don't end when someone gives you a recommendation. They continue with questions like:

What am I actually deciding?

What am I assuming?

What would change the answer?

What information is actually worth getting?

Am I repeating a pattern from previous decisions?

At what point have I resolved enough uncertainty to act?

That became the distinction I kept seeing.


The process difference

Same starting point. Different number of steps before an answer becomes trustworthy.

General-purpose AI
Decision
Analysis
Recommendation
Quorum
Decision
Structural read
Council
Challenge
Update
Conditions
Next action
Decision record

The difference isn't necessarily the answer. It's the decision process around the answer.

See how each step of Quorum's process works →

Decision 1

₹48L job or ₹34L stable job?

Career · Money · Family · Risk

A 34-year-old, married, with a young child, is choosing between the current ₹34L stable, predictable role and a new ₹48L role with higher visibility and upside but less predictability. The household has approximately ₹2Cr invested and ₹53L of home debt.

Gemini

Gemini framed the decision as career acceleration vs family stability and psychological comfort.

It recommended taking the ₹48L opportunity, provided the new company and manager checked out.

Strong answer.

Quorum

Quorum also leaned toward taking the role initially. But it focused more heavily on what needed to be resolved before accepting it:

  • household downside
  • company stability
  • role clarity
  • exit optionality
  • whether the role actually fits the career identity being built

Then we challenged the council. The recommendation moved toward Wait.

Not because the ₹48L role suddenly became bad. Because the council concluded that moving before resolving the structural uncertainties was the bigger mistake.

The interesting difference
The question changed from:
Should I take the job?
What has to be true before taking the job becomes a decision I can trust?

Decision 2

Buy the ₹2.5Cr home or keep investing?

Money · Lifestyle · Family · Optionality

A 34-year-old family has approximately ₹2Cr invested, spends around ₹1.8L/month, invests roughly ₹2.5–2.6L/month, and is considering a ₹2.5Cr home.

Gemini

Gemini's strongest framing was current lifestyle consumption vs early financial autonomy.

It recommended waiting roughly three years so the financial corpus could compound further before making the purchase.

Again, a strong answer.

Quorum

Quorum also said Wait. But instead of treating the decision purely as rent-versus-buy economics, it asked whether the home was actually solving something else: control? stability? family quality of life? a desire to feel settled?

It also highlighted the opportunity cost of permanently compressing the household's investable surplus.

Then we challenged it. The council moved from Wait to Mixed. Five advisors changed position.

The key question became: what is the actual post-EMI monthly investable surplus, and does the family's FI trajectory still survive it? The decision became conditional.

Wait challenge Mixed
5 advisors shifted

Decision 3

Move abroad or stay in India?

Career · Family · Identity · Geography
Gemini

Gemini's recommendation was essentially: stay in India, protect the family's financial trajectory and support network, and find international exposure without permanent relocation.

Its major concerns were:

  • spouse employability
  • childcare costs
  • loss of family support
  • financial drag
Quorum

Quorum also initially leaned toward waiting. But its question was: are you treating this as a career decision when the entire household bears the risk?

It highlighted:

  • spouse career continuity
  • work authorization
  • household cash-flow resilience
  • the cost of returning
  • the possibility that "going abroad" is partly an identity/status aspiration

Then we challenged the council. The overall recommendation stayed Wait, but internally the council became much less one-sided. One advisor moved all the way from Wait to Proceed. Others moved to Mixed.

Wait challenge Wait
Overall verdict held — but internally, 1 advisor moved all the way to Proceed and others moved to Mixed

The important thing wasn't that Quorum changed the final answer. It was that the reasoning state changed.


Decision 4

Keep the job or build a company?

Entrepreneurship · Identity · Wealth · Regret
Gemini

Gemini strongly recommended: do not quit immediately. Run a 6-month staged validation while employed. That means:

  • customer discovery
  • paid pilots
  • evidence of willingness to pay
  • predefined quit criteria

This was excellent.

Quorum

Quorum landed in almost exactly the same practical place. But it added a personal layer: you may be changing the threshold you previously said you required because you want entrepreneurship badly enough.

"You set the threshold, then move the threshold when you meet it."

That's not startup advice. It is feedback about the decision-maker.

Then we challenged the council and made it argue for quitting now. The council still recommended staged validation, but the internal council became more mixed.

What stood out
A decision system can challenge the decision-maker, not just the decision.

Decision 5

What should a family actually optimise for?

Family · Marriage · Career · Money · Optionality
Gemini

Gemini argued for a version of family bandwidth + career leverage + financial security, without aggressively maximising wealth or permanently relocating.

Its key idea was that the marginal value of additional wealth may be lower than the value of parental presence, marriage bandwidth and a relaxed family environment during the child's early years.

Quorum

Quorum reframed the problem as: should this family lock itself into one dominant life architecture now, or deliberately buy the next two years as an option-creation period?

Its initial recommendation was staged optionality. We challenged that recommendation directly. This time the council actually shifted its verdict: Mixed → Proceed with the staged / optionality-first path. Two advisors changed from Mixed to Proceed.

And the conditions became very concrete:

  • both spouses must genuinely co-author the plan
  • career "push windows" need actual time limits
  • no irreversible geographic/career commitment before a 24-month review
  • international exposure should first be tested as a reversible experiment
  • the plan must have exit criteria, not vague promises of optionality
Mixed challenge Proceed
2 advisors shifted · staged / optionality-first path
"Vague optionality is not optionality; it is drift with a better story."

I didn't learn Quorum is smarter than Gemini. That would be an easy claim and an uninteresting one. Gemini was very good.

General-purpose AI is becoming very good at answering "What should I do?"

But difficult decisions often require a second layer: "What should I resolve before I trust that answer?"

That is what Quorum is designed around.


The difference in one picture

One more look at the shape of it.

General-purpose AI
Decision
Analysis → Recommendation
Quorum
Decision
Structural read → Council → Challenge → Update → Conditions → Next action → Decision record

And because Quorum keeps a structured record, there is another layer over time:

How do I tend to make decisions?

Not generic personality advice. Your own recurring patterns — the layer Mirror is built around.


Recurring behaviors, not proven constructs

The things Quorum kept looking for.

Hidden variable

The thing that could actually change the answer.

Irreversible commitment

The point after which reversing becomes expensive.

Missing evidence

The information worth getting before deciding.

Third option

A way to reduce uncertainty without immediately committing to A or B.

Decision pattern

The way this person has made similar choices before.


Maybe the problem isn't "better AI advice"

AI has made answers cheap. The scarce resource is knowing what to trust, what to question, and when to act. That's why I built Quorum.

It doesn't start by asking:

"What should you do?"

It starts by asking:

"What are you actually deciding?"

And sometimes the answer is:

You're not ready to decide yet.


Try it with your own decision

Have something consequential you're genuinely stuck on?

Don't give Quorum a hypothetical. Give it the real thing.


Methodology

How we ran the experiment

Five simulated but realistic decision scenarios were run through both a general-purpose AI and Quorum using the same underlying information. For each scenario:

  1. The decision was given to both systems.
  2. The initial recommendation was recorded.
  3. Quorum was deliberately challenged and asked to prove its initial position wrong.
  4. We recorded whether the council changed position, which advisors moved, and what condition became decisive.
  5. The comparison focuses on reasoning process and product behavior, not on claiming that one model is objectively more intelligent.

The scenarios are illustrative/composite and do not represent published private decisions from other people.