The Seven-Month Wall: Why AI Alone Can’t Build What I’m Building—And What Comes Next

This isn’t just a technical puzzle. It’s the culmination of a lifetime’s work. And no, the big banks can’t just copy it.

Bill Cara

January 19, 2026

For the past seven months, I’ve been chasing a vision that feels almost alchemical: turn one massive, daily financial dataset into five distinct, polished reports—three daily, two weekly—through AI alone. No manual reshaping, no endless Excel gymnastics. Just a prompt, a process, and publication.

It was meant to be elegant. A single AI platform would ingest well over 2,000 rows of data, apply a complex, homegrown schema, and output perfectly formatted tables and nuanced commentary, ready for my subscribers. I called it “The Navigator” as a service to investors who understand they need decision-support or would be comforted by it. For months, I refined prompts, tweaked mappings, and pushed platforms to their breaking points, searching for that elusive all-in-one engine.

Last night, I hit the brick wall. I know it and admit it. And the clearest diagnosis came from the very platform that came closest: Perplexity.

But before I share the technical issues and solution, let me be clear about something: this project was never just about software. The copycats who might see today’s review and think they’ve found a shortcut are missing the point entirely. You can’t replicate the why, or the how, without the foundation that took a lifetime to build.

The Foundation You Can’t Prompt-Engineer

What powers the system I am building isn’t just clever code. It’s ten years in chartered accounting and management consulting, learning how systems and risk actually work. It’s five subsequent decades in the securities industry, registered and operating in three different countries, from the trading floor to portfolio management to the boardroom. It’s the visceral understanding of market mechanics, regulatory nuance, and, most importantly, the structural failures of the industry itself.

I’ve worked at the most senior levels inside the major financial institutions. I know exactly why they will never build what I’m building. The buy-side vs. sell-side conflicts paralyze objective analysis. Organizational pressures and internal competition among A-type personalities fracture focus and incentivize siloed, politically-safe work, not transformative tools for end investors. They can’t do it. They won’t even try.

Smaller firms? They’re either too busy chasing career milestones or drowning in committee processes that birth three-humped camels—expensive, clumsy, and unfit for purpose.

So, while someone might copy a technical workflow, they cannot replicate the decades of context, the mission, or the uncompromised perspective that informs every line of every report. This isn’t arrogance; it’s geometry. You can’t take a shortcut to a vantage point that requires a 50-year climb.

The Core Problem: AI is a Brilliant Analyst, Not a Deterministic Database

The reply from Perplexity to me today was a masterclass in clarity. It confirmed my sinking feeling: the problem isn’t the AI’s intelligence—it’s its role.

Perplexity excels at analysis: synthesizing trends, writing sharp commentary, connecting cross-currents. Where it fails—catastrophically—is in guaranteed, lossless data structuring. For a report demanding 88 precise tables with strict row ranges (R-01 through R-88), no omissions, and perfectly repeated formats, a chat interface is a liability. It might hallucinate a row. It might skip one. The layout might crumble. In financial reporting, that’s not an error; it’s a breach of trust. I operate on trust principles and a ‘no garbage’ policy.

As the response put it: *“Where it will fail is if you ask it, in one shot, to: ‘scan all R-01–R-88 instruments, derive tables, and write full commentary.’”* That was exactly my “Navigator failure mode.”

The Revealed Compromise: A Two-Engine Architecture

The solution isn’t to abandon the AI that finally understood my ask. It’s to stop asking it to do something it wasn’t built for. The recommended path is a division of labor:

1. The Deterministic Data Engine: This is the new backbone. Use a tool built for rigorous data manipulation—ChatGPT’s Advanced Data Analysis (which runs Python/Pandas), Excel-native AI (Copilot, Excelmatic), or even a lightweight database manager. This engine’s sole job is to ingest the raw daily dataset, apply the schema flawlessly, and generate every required table. It does the mechanical, zero-tolerance work. This is where my proprietary logic—the “secret sauce” forged from that decades-long career—resides securely.

2. The Narrative Intelligence Engine: This is where Perplexity stays. Its new job is purely analytical: read the pre-built, curated tables from Engine 1, and generate the insightful commentary, the sleeve-level summaries, and the macro risk assessments. This is its sweet spot.

The Mission Behind the Machinery

At 2:30 am today I asked myself, why go through this hell? It’s not for a technical trophy. But I published what I had to continue my journey. I am not a quitter.
My mission is to democratize the role of securities investment by putting decisions squarely in the hands of the owners of wealth. To bypass the advisor-manager-banker complex that thrives on opacity and conflict. My reports are designed to be informative, educational, and actionable—giving capital owners the clarity and confidence to engage directly with their own financial destinies.

This two-engine architecture I will now pursue serves that mission. It ensures bulletproof reliability (non-negotiable for trust) while freeing the analytical voice to educate and empower. It modularizes the system so the invaluable logic is protected, while the output remains powerful and clear.

The Path Forward

So, the seven-month journey wasn’t a failure. It was an incredibly expensive—in hours, not dollars—proof of concept. It proved that:

• AI is ready to be the voice of sophisticated reporting.

• But it cannot yet be its spine for highly structured data. Software is needed – no matter the current narrative you are being exposed to.

• The hybrid future is here: AI + deterministic data tooling.

My next phase is building this pipeline. The goal remains: one daily data dump, five automated reports. But now, the responsibility is rightly distributed. Let the code do what it does best: be exact. Let the language model do what it does best: find meaning and tell the story.

The wall I ran into after midnight last night wasn’t the end. It was just the blueprint for a bridge -- a bridge that can’t be crossed without the right foundation.

Today, I’ll see what I can produce for the Cara Playbook. It might be sample #2. I’ll also produce the next Cara Portfolio Assessment Report, which I know I can do separately from the other reports. I will make all of this content available free to all subscribers and for those who have paid money, I have paused the billing. You might wish to pledge money on one of the premium sites (caraportfolio.substack.com, caraplaybook.substack.com, or INSTAT.substack.com) to show me you care about my mission’s success, but as always that is a personal decision.

P.S. For those on a similar path: the most valuable lesson I’ve learned in the past few months wasn’t technical. It was learning to diagnose what kind of problem we’re actually solving. Is it a data structuring problem, or a data storytelling problem? Despite the promises, very few tools are good at both. And even fewer people have the background to know the difference, or the mission that makes the struggle worthwhile.