Friday, March 27, 2026

Everything You Need to Know About AI Is Already Spinning at the Fair

The Mindful Dollar  ·  AI Explained Simply

Everything You Need to Know About AI Is Already Spinning at the Fair

You don't need a computer science degree to understand artificial intelligence. You just need to have spent an afternoon at the youth fair.

Beginner AI Explained Real World

Today I found myself standing at the youth fair, watching my daughter ride the her favorite ride for the third time in a row. And somewhere between the cotton candy and the music box melody, something clicked.

I spend a lot of time explaining AI to people, all business owners, creatives, people who feel like the whole thing is some kind of magic trick happening behind a curtain they're not allowed to peek behind. And I realized: the rides at the fair are the most honest metaphors I've ever seen for how AI actually works.

Not just the carousel. All of them.

So let's walk the fairgrounds together.

First — what even is AI?

At the most basic level, artificial intelligence is a system that learns from examples in order to make predictions or decisions. It doesn't think the way you think. It doesn't have feelings. It doesn't know what a horse is the way your brain knows what a horse is.

What it does is find patterns. Lots and lots of patterns. And then it uses those patterns to figure out what comes next.

That's it. Everything else — ChatGPT, image generators, recommendation engines, the thing that knows you want to watch another episode — is built on top of that simple idea.

Start here: the carousel

The carousel is the best place to begin because it shows you the whole engine at once.

A carousel has horses mounted on poles, attached to a spinning platform. Each horse moves in the same circular path, but the poles make them rise and fall at slightly different times. No horse goes rogue. No horse decides to gallop off into the crowd. They all follow the same underlying rhythm — controlled by a mechanical system they didn't choose and don't understand.

The platform is the training data

The spinning platform everything is attached to? That's your training data. It's the foundation — every book, article, conversation, and record the AI was exposed to before it ever talked to you. Just like you can't have a carousel without the platform, you can't have an AI without the data it learned from.

The horses are the patterns

Each horse represents a pattern the AI has learned. Some are tall and obvious — massive patterns the system sees constantly. Others are small and subtle. That variation is what allows the system to handle nuance — to know that "bank" means something different beside a river than beside an ATM.

The poles are the weights

The poles control how high or low each horse goes. In AI, these are called weights — numbers that tell the system how much importance to give any pattern. When an AI is being trained, it's constantly adjusting those poles. A horse that kept predicting wrong gets its pole shortened. One that was consistently right gets extended. That process of adjustment is called learning. It's not magic. It's math.

The music is the prompt

The carousel only runs when the music plays. Your question — the thing you type into the AI — is that music. The platform doesn't change. The horses don't change. But the specific ride you get depends entirely on the song you start playing. This is why how you ask matters just as much as what you ask.

AI isn't smarter than you. It has seen more examples than you. Those are two very different things — and the difference matters.

Now walk the rest of the fairgrounds

Once you understand the carousel, every other ride at the fair starts to look familiar. Here's what the rest of them are trying to tell you.

🎢
Roller Coaster
How AI learns: gradient descent

The car climbs to the top of a hill (high error) and rolls down into a valley (low error). AI training literally follows this same logic — it keeps rolling downhill toward the lowest possible mistake. The loop-de-loops? Those are the unexpected twists in real-world data.

🌀
The Scrambler
How AI runs fast: parallel processing

Multiple arms spinning at once, each car moving independently but governed by the same central motor. That's exactly how AI runs — thousands of calculations happening in parallel, all coordinated by the same underlying system. One engine, many simultaneous riders.

🎡
Ferris Wheel
How AI gets smarter: training loops

The wheel rotates through the same path over and over, gaining a slightly different perspective each revolution. AI models do the same — they cycle through training data repeatedly, adjusting their understanding a little more each pass until performance peaks.

🎠
The Carousel (again)
How AI recognizes things: pattern recognition

Every horse moves up and down in a predictable rhythm. AI pattern recognition works the same way — it identifies repeating structures in data and uses them to predict what belongs next. Familiar motion, reliable prediction.

🌪️
The Tilt-A-Whirl
How AI surprises you: emergent behavior

The cars spin unpredictably even though the physics are entirely deterministic. This mirrors how large AI models produce outputs that feel creative or surprising — even though they're running on pure math. Small changes in input can produce wildly different rides.

🎟️
The Ride Operator
How AI gets shaped: the system prompt

The operator controls the speed, the duration, and who gets on. You never see them — you're just on the ride. Behind every AI product there's a hidden set of instructions from the company that shapes how it behaves. Same machine, very different experiences depending on who's running it.

So what does this all mean for you?

Understanding the fair changes your relationship to the rides.

When you know that AI is pattern-matching from a platform of data — not thinking, not feeling, not reasoning like a human — you stop being mystified by it. You start asking better questions. You notice when the output seems off because the training data probably didn't include enough of your context. You realize the tool is only as good as the music you give it.

You also stop being afraid of it. A carousel is a powerful, intricate machine. It can be dangerous if misused. But once you understand how it works, you can get on and enjoy the ride — or decide when to step off.

Every ride at the fair was built by people. Designed, tested, adjusted. AI is no different. And just like no one feels embarrassed for not knowing how a Ferris wheel motor works, you don't need to understand the math behind a neural network to use these tools wisely.

You just need to know enough to stay in your seat and hold on.

The carousel doesn't know it's a carousel. The horses don't know they're horses. The whole beautiful, spinning thing is a system — designed by people, built from data, and set in motion by you every time you ask it a question.

The one thing to remember

AI isn't magic. It's a very sophisticated fair.

Some rides are thrilling. Some are disorienting. Some are built better than others. And there will always be a new one going up next season that everyone says you have to try.

But now you know how the machines work. And that changes everything.

Pick your horse. The ride is yours.


Questions? Reactions? Did this land differently than you expected? Hit reply — I read everything. And if you know someone who's been too intimidated to even start exploring AI, send them this one. Sometimes all it takes is the right analogy.

Monday, March 23, 2026

"Skilled Labor Is Dead." I Disagree

 The Mindful Dollar  ·  AI Explained Simply

"Skilled Labor Is Dead." I Disagree.

By Nasly Duarte | Mindful Dollar — Doing More With Less

I was sitting in a virtual live event this week when the speaker put up a pyramid. Tech at the top. Media and Data below it. Intellectual Property in the middle. And near the bottom, in red text: Skilled Labor. Component Labor at the base.

Then the speaker said it: "Skilled labor is dead."

The chat started moving. People nodding along. The room was full of builders and tech-forward thinkers ready to automate the future from behind a screen.

And I thought: that's not just wrong. It's incomplete in a way that could actually hurt people.

I've spent 20 years in accounting. I've lived the screen-only career. And what I've learned — through my body, not just my brain — is that the conversation about the future of work is missing something critical.

The Pyramid Gets It Backwards

The U.S. construction industry needs to attract 349,000 new workers in 2026 alone, according to Associated Builders and Contractors. Over the next decade, the industry will need 1.9 million workers just to keep up with growth and retirements. And 91% of construction firms report struggling to find qualified workers.

The retirement cliff is staggering: 41% of the construction workforce will retire by 2031. For every five Baby Boomers leaving the trades, only two younger workers are coming in behind them.

This isn't just construction. In manufacturing, 2.1 million jobs could go unfilled by 2030. And 77% of manufacturers report ongoing difficulty even finding workers.

Meanwhile — and this is the part that should make anyone in the tech space pay attention — white-collar layoffs have dominated headlines in 2025 and 2026, with technology, media, and finance companies cutting tens of thousands of positions. The skilled trades are experiencing the exact opposite. Demand is outstripping supply.

Every piece of AI technology we build still needs physical infrastructure. Servers need buildings. Buildings need electricians. Data centers need cooling systems. The cloud lives in a warehouse somewhere, and somebody had to pour that concrete, run that wiring, and connect those pipes.

When I hear "skilled labor is dead," I hear someone who has never had to call a plumber on a Sunday.

But that's not actually the argument I want to make. The real issue goes deeper.

What 20 Years at a Desk Taught Me That No Conference Will

Here's what nobody warned me about when I started my accounting career: sitting at a desk for hours and hours is a health hazard.

It doesn't matter if you work four-day weeks or seven-day weeks. It doesn't matter if AI cuts your workload to four hours a day. If you're sitting for long stretches, your body is paying a price. Slowly. Quietly. Until it isn't quiet anymore.

I lived it. Years of sedentary work did real damage to my body — damage I didn't see coming until it was already done. If you want the full story, follow my Think Like a Healer series, where I document what years of sedentary work actually does and what I'm doing to reverse it.

That experience gave me a perspective on the future of work that I don't hear anyone in the AI space talking about.

The Balance Problem

Tech work is a lot like accounting work. It takes enormous brain power. Hours of deep focus. Mental stamina. You're solving complex problems, holding abstract systems in your head, making decisions that ripple downstream.

But the human brain doesn't work in isolation from the body.

Even if we build the most incredible AI tools. Even if we automate every workflow. Even if we can run a business in four hours a day from a laptop — we still need to develop both our mind and body. Not as a nice-to-have. As a biological requirement.

Your brain chemistry needs physical output. Not just for fitness. For clarity. For regulation. For the kind of deep creative thinking that no amount of screen time produces on its own.

You can't optimize your way out of that with a better prompt.

I call this the Balance Problem, and I think it's the blind spot in every conversation about the future of AI and work.

The Interchange

Here's what I think is actually coming over the next five to ten years. I'm calling it The Interchange — the convergence of tech workers picking up trade skills and trade workers picking up tech skills. Not because the market forces them to. Because their bodies and brains require it.

The Claude builder who automates their entire content pipeline will pick up woodworking, or gardening, or welding — because they realize their sharpest thinking happens after they've used their hands.

The electrician who runs a crew of ten will start using AI to handle estimates, scheduling, and accounting — because they realize the administrative burden is what's actually burning them out.

This reminds me of the Netflix documentary The Biggest Little Farm, where John and Molly Chester buy a barren plot of land outside Los Angeles and spend eight years turning it into a thriving, biodiverse farm. The whole lesson of that film is that the ecosystem balances itself out — but only when you let both sides exist. The soil needs decomposition and growth. The land needs predators and prey. Nothing thrives in isolation.

AI — my hope — will balance us the same way. Not replace skilled labor. Balance us.

That's why I'm building Mindful Dollar at the intersection of both. I'm studying Applied AI and building software architecture — and I'm also listening to my body after two decades of desk work. "Doing More With Less" isn't just about efficiency. It's about building a life that doesn't break your body to feed your brain.

Where Do You Stand?

I wrote this post because I couldn't let "skilled labor is dead" sit unchallenged.

Are you a skilled trade and tech combo person? A developer who's picked up a physical craft? A contractor who's learning to automate? Are you interested in using AI not just to build faster, but to build a more balanced life?

Where do you see trade skills going — not just in Claude, but in the real world?

Drop a comment, send me a message, connect with me on LinkedIn. The best responses will show up in a follow-up post, because this conversation is bigger than one article.

We're building in public. Let's figure out what we're actually building toward.

Follow the Think Like a Healer series for more on what years of sedentary work does to your body — and how to reverse it.

Mindful Dollar | Nasly Duarte | Doing More With Less

#BuildInPublic #SkilledTrades #AI #TechAndTrades #MindfulDollar #DoingMoreWithLess #ThinkLikeAHealer #FutureOfWork #BalanceProblem

Sources

  • Associated Builders and Contractors (ABC) — "Construction industry must attract 349,000 workers in 2026" — abc.org
  • Associated General Contractors of America — 91% of firms struggle to find qualified workers
  • National Association of Home Builders (NAHB) — 41% of construction workforce will retire by 2031; housing industry labor shortage carries $10.8 billion annual economic impact — nahb.org
  • Deloitte & The Manufacturing Institute — 2.1 million manufacturing jobs could go unfilled by 2030; 77% of manufacturers report difficulty attracting workers — deloitte.com
  • Academy of Craft Training — Construction will need 1.9 million workers over the next decade — academyofcrafttraining.org
  • Skillwork — For every 5 Baby Boomer retirees, only 2 younger workers enter the trades — skillwork.com
  • Metaintro — White-collar layoffs in tech/media/finance vs. trades demand outstripping supply, 2025–2026 — metaintro.com
  • U.S. Chamber of Commerce — America Works Data Center, industry labor shortage data — uschamber.com
  • The Biggest Little Farm (2018) — Directed by John Chester. Available on Netflix, Hulu, and Amazon Prime Video.

Sunday, March 22, 2026

Is n8n Dead for Content Workflows?

The Mindful Dollar  ·  AI Explained Simply

Google Just Put an AI Content Engine Inside Your Spreadsheet. Is n8n Dead for Content Workflows?

By Nasly Duarte | Mindful Dollar — Doing More With Less


I need to start with a confession: I almost mass-deleted three n8n workflows this week.

I was organizing my LinkedIn content topics in Google Sheets — hashtags, hooks, post drafts, status columns — and then I saw it. A little purple cross icon appeared with a label I hadn't noticed before: "Drag to fill with Gemini."

I dragged it. And Gemini started filling my rows with content. Not random content. Content that matched my column headers and the patterns in my existing rows. It pulled from the web. It categorized. It wrote draft hooks based on the topics I'd already started.

I sat there staring at it like… wait. Did Google just eliminate half my automation stack?

I want to know if you had the same reaction — or if you think I'm overreacting. Keep reading and tell me at the end.

What Actually Changed (This Literally Just Happened)

Google rolled out a major Gemini update to Workspace on March 19, 2026. The Sheets update is the one that matters for builders like us who plan content in spreadsheets.

Here's what "Fill with Gemini" actually does:

  • You set up column headers (Topic, Hook, Hashtags, Post Draft, Status)
  • You fill in a few rows manually so Gemini can see the pattern
  • You drag down and Gemini auto-populates the rest — pulling from the web, categorizing, even writing draft copy

It's not just autocomplete. It runs a separate web search for each row. It reads your column headers and figures out what data to pull. Google's VP of Product described it as Gemini being able to "figure out how to go find what you need" just by reading the structure you've built.

Available now for Google AI Ultra and Pro subscribers.

My Hot Take: Gemini Replaces the Brain. n8n Is Still the Body.

What Gemini in Sheets Does Well

Topic research and enrichment. If I have a column of broad topics — "Miami construction workforce," "AI in accounting," "contractor licensing Florida" — Gemini fills adjacent columns with trending angles, relevant statistics, and draft hooks. That used to require a separate n8n node hitting an AI API.

Pattern-based content generation. Write three posts in a specific format, drag down, and Gemini generates more in that same pattern. For batching a week's worth of content in one sitting, this is fast.

Zero setup cost. No webhook URLs, no OAuth tokens, no node configuration. It just works inside the sheet you already have open.

What Gemini in Sheets Cannot Do

It cannot publish. Gemini fills your spreadsheet. It does not post to LinkedIn. You still need either a scheduling tool or an automation platform to move content from sheet to platform.

It cannot run on a schedule. Gemini responds when you interact with it. There's no "run this every Monday at 6 AM" trigger. n8n's Schedule Trigger exists specifically for this.

It cannot handle approval workflows. If you want a human review step before publishing (and you should — more on this below), Gemini doesn't offer that. n8n can route a draft to Telegram or Slack for your sign-off before it goes live.

It doesn't close the loop. After a post publishes, n8n can mark the row as "Posted" and add the live URL back to your sheet. Gemini doesn't do post-publish tracking.


Here's Where I Want to Fight About It

I think most builders are overengineering the research step.

If you're paying for Google AI Pro anyway, why are you running a 10-node n8n workflow with Perplexity and OpenAI just to generate topic ideas and draft hooks? Gemini does this natively now, inside the sheet, with zero config.

But — and this is important — I also think anyone publishing AI-generated content straight to LinkedIn without human review is playing with fire.

Where do you land on this?

  • Are you comfortable with fully automated post-to-publish pipelines?
  • Or do you keep a "Ready" column and review every draft before it goes out?
  • Has an AI-generated post ever embarrassed you? (Be honest.)

I keep a human-in-the-loop step because my voice is my brand. But I know builders who skip it entirely and post three times a day on autopilot. I want to hear what's actually working for you.

The Stack I'm Landing On (Tear It Apart)

Here's my current thinking. Tell me what you'd change:

Layer Tool Why
Topic research & draft generation Gemini in Sheets Faster, zero config, web-grounded
Human review & editing Me, in the same Sheet Edit Gemini's drafts, mark rows as "Ready"
Scheduled publishing n8n Reads "Ready" rows, posts to LinkedIn on a schedule
Status tracking n8n Updates row to "Posted," adds the live post URL

Gemini replaced the research and first-draft nodes in my n8n workflow. It did NOT replace the scheduling, publishing, and status-tracking nodes.

Gemini is the writer. n8n is the operations manager.

Less nodes. Less complexity. Same output. Maybe better output, because Gemini's web search is pulling fresher data than a cached AI prompt would.


The Bigger Pattern (This Isn't Just About Content)

This is something I keep seeing as AI tools mature, and I think it matters for every builder reading this:

The generation layer is getting commoditized. Every platform is adding AI content generation natively — Google Sheets, Notion, Canva, even email clients. The thing that still requires you to build something is the orchestration layer — getting the right content to the right place at the right time with the right approvals.

That's where n8n (or Zapier, or Make, or Claude Code) still earns its keep.

But here's my question for the group: how long until Google adds scheduling and auto-publish to Sheets too? If Gemini can already research, draft, and organize — publishing feels like it's next. And if that happens, does n8n lose its role in this workflow entirely?

I genuinely don't know the answer. I'm curious what you think.

What Should I Doing Next (And What I Want From You)

Should I simplify my n8n workflow this week. Remove the OpenAI research nodes and let Gemini handle topic enrichment directly in my content calendar sheet. My n8n setup will get leaner — it just reads, publishes, and updates.

Doing more with less. That's the whole point.

Next week, I'm building my full n8n workflow setup — screenshots, node configs, and the Google Sheet template. So you can use it or tell me what's wrong with it.

But before that, I want to hear from you. Seriously. Drop a comment, reply, DM me — whatever works:

  1. What does your content automation stack look like right now? (Sheet + n8n? Notion + Zapier? Something else entirely?)
  2. Have you tried "Fill with Gemini" yet? What was your first reaction?
  3. Human review or full autopilot? And has skipping review ever burned you?
  4. Hot take: will Google eventually add scheduling and publishing directly to Sheets? Or is that a bridge too far for a spreadsheet app?

The best answers are going in next week's post. I'll credit you and link your profile. Build in public means building together.


Connect with me on LinkedIn — I'm documenting this entire workflow evolution .

Mindful Dollar | Nasly Duarte | Doing More With Less


#BuildInPublic #n8n #GeminiAI #GoogleSheets #ContentAutomation #LinkedInStrategy #MindfulDollar #DoingMoreWithLess

Tuesday, March 10, 2026

What If We Could Map the Gap Between What Companies Need and What Schools Teach?

The Mindful Dollar  ·  AI Explained Simply 

What If We Could Map the Gap Between What Companies Need and What Schools Teach?

I’m building a workforce intelligence dataset to prove what everyone feels but nobody is measuring.

By Nasly Duarte

Calling It Out Isn’t Enough

In my last post, I broke down a job listing that asked one person to do the work of three — an Assistant Controller role carrying Controller and CFO responsibilities at entry-level pay. The response was overwhelming. Employees said that’s my job right now. Business owners said I didn’t realize I was doing that. Educators said we don’t have a way to see this.

That last response is the one that kept me up at night.

Because I’ve been having this conversation for a while now. I’ve spoken to workforce development organizations, school directors, program administrators, nonprofit leaders, and economic development professionals in Miami about this exact structural issue. The response is usually some version of acknowledgment followed by inaction — not because people don’t care, but because they don’t have the data to justify changing anything.

Anecdotes aren’t enough. Frustration isn’t a metric. If I want institutions to move, I need to give them something they can measure, present to a board, and build a curriculum around.

So I’m building it.

What Cybersecurity Taught Me About Workforce Intelligence

In cybersecurity, there’s a company called SpyCloud. They don’t wait for data breaches to happen. They continuously scrape the dark web, structure the raw data into threat intelligence, and deliver it to organizations so they can act before the damage hits. They turned chaos into a dataset, and that dataset became a product that protects millions of people.

That model stopped me in my tracks. Because the workforce pipeline has the same problem — just in a different domain.

Companies are posting job descriptions that reveal exactly where their operations are broken. That data is sitting in plain sight on Indeed, LinkedIn, and government job boards. Every single day, employers are telling us what they need — and if we look closely enough, they’re also telling us what they don’t understand about what they need.

Meanwhile, schools are building curricula in silos. An accounting program teaches accounting. A project management program teaches project management. An IT program teaches IT. But the job market isn’t hiring in silos anymore — companies want people who can operate across all three. And nobody is measuring that mismatch at the local level where it matters most.

What if we could do for workforce development what SpyCloud does for cybersecurity? What if we could scrape, structure, and analyze job postings to create a real-time intelligence feed that shows schools, workforce programs, and economic development offices exactly where the gaps are?

The Gap Nobody Is Measuring

Let me be clear about what doesn’t exist right now.

The Bureau of Labor Statistics tracks broad occupational categories. LinkedIn publishes national hiring trends. The Department of Economic Opportunity releases workforce data at the state level. These are useful for understanding the macro picture.

But none of them do what I’m describing.

Nobody is taking actual job postings from South Florida employers — the ones posted this week, this month — and breaking them apart responsibility by responsibility. Nobody is counting how many distinct roles are compressed into a single title. Nobody is comparing the required skill combinations against what local programs actually teach. Nobody is scoring the alignment between what employers demand and what graduates are prepared to do.

Not in Miami. Not anywhere that’s accessible to the people making curriculum decisions at the institutions that feed our local workforce.

This means program directors are designing courses based on industry standards from five years ago, advisory board feedback that comes once a quarter, and their own professional experience — which may or may not reflect what’s happening on the ground right now. They’re building planes without radar.

What This Kind of Analysis Would Actually Show

Let me paint a picture using the job post I analyzed in my last blog as a starting point.

That one posting contained responsibilities spanning three distinct finance roles. It required proficiency across four different software platforms. It listed 16 separate duties. The salary topped out at $90,000 for work that, properly distributed, would cost a company $275,000 or more in combined headcount.

Now imagine running that same analysis across every accounting, finance, and operations job posted in Miami-Dade, Broward, and Palm Beach counties over the last 12 months. The dataset could surface patterns like:

Metric

What It Reveals

Role Compression Ratio

How many distinct roles are packed into a single title. A ratio of 3:1 means one person is expected to do three jobs.

Skill Mismatch Score

The percentage of required skills in a job post that are not covered by the corresponding local degree or certificate program.

Salary-to-Scope Alignment

Whether the pay offered matches the actual scope of responsibilities — or whether employers are buying three roles at one role’s price.

Curriculum Coverage Gap

The specific skills and competencies that appear frequently in job postings but are absent from local program curricula.

Technology Stack Drift

The software and platforms employers require versus what schools train students on — exposing how far behind training has fallen.

These aren’t opinions. They’re measurable, repeatable data points. And they give decision-makers something they’ve never had: a real-time map of the gap between the workforce pipeline and the workforce reality.

How You Actually Build This

I’m an AI student at Miami Dade College studying Computer Vision and Natural Language Processing. So let me walk you through what this looks like under the hood — in plain language, because the people who need this tool aren’t engineers. They’re educators, workforce developers, and business leaders.

Step 1: Collection. A web scraper pulls job postings from Indeed, LinkedIn, and government job boards, filtered to Miami-Dade, Broward, and Palm Beach counties. It collects the title, salary range, responsibilities, required skills, preferred qualifications, and industry. This runs continuously, building a living dataset that grows every day.

Step 2: Classification. A Natural Language Processing pipeline reads each job post and extracts the individual responsibilities, required skills, and software requirements. It doesn’t just count keywords — it understands context. It knows that “support the annual budgeting process” is a CFO-level responsibility, not an admin task. It classifies each responsibility by role tier and skill domain.

Step 3: Scoring. A comparison model maps the extracted skills and responsibilities against published curricula from local institutions — what MDC teaches in its accounting program, what Miami Tech Works covers in its workforce training, what certificate programs include in their course outlines. The model generates a coverage score: what percentage of what employers are asking for is actually being taught?

Step 4: Visualization. A dashboard presents the findings in a format that non-technical stakeholders can act on. Program directors see which skills are missing from their curricula. Workforce developers see which industries have the widest gaps. Economic development offices see where training investments would have the highest return. Employers see how their job posts compare to market norms.

This isn’t theoretical. Every component I just described uses technology I’m already working with in my AI program. The NLP pipeline, the classification models, the data architecture — these are the same tools I’ve been building with. The difference is the application. Instead of analyzing health data, I’m analyzing the health of the workforce pipeline.

Who Gets What From This

One dataset. Multiple stakeholders. Each one gets something different.

Schools and colleges get a real-time curriculum audit. Instead of relying on advisory boards that meet twice a year, program directors can see which skills are appearing in job postings right now and compare that against what they’re teaching this semester. They can identify gaps before students graduate into them.

Workforce development programs get targeting precision. Instead of broad assumptions about what industries need, they can see exactly which skill combinations are in demand and build training programs that match the market — not the market from three years ago, but the market from this month.

Economic development offices get ROI data for training investments. If the dataset shows that 80% of construction accounting roles in South Florida require ERP integration skills and zero local programs teach it, that’s a clear, fundable gap. It turns “we need more workforce development” into “we need this specific training program and here’s the data proving demand.”

Employers get a mirror. The dataset can show them that their job post is compressing three roles into one, that their salary is misaligned with their scope, or that the skills they’re requiring don’t exist in the local talent pool because nobody is training for them. Some won’t want to hear it. The best ones will use it to restructure their hiring and finally build the systems they’ve been trying to replace with people.

Students and job seekers get visibility. They can see what the market actually demands before they invest two years in a program. They can identify which additional skills would make them competitive. And they can walk into interviews with a clearer understanding of whether a role is structured fairly or whether they’re being asked to carry a department on their back.

I’m Building This in Public. And I Want You in the Room.

I’m not waiting for permission. I’m not waiting for a grant. I’m not waiting for someone to greenlight a study. I’ve been talking to organizations about this problem for long enough to know that the conversation alone doesn’t move the needle. Data does.

So I’m building this. I’m documenting the process. And I’m sharing it openly because the only way this works is if the people who need it can see it, challenge it, and shape it alongside me.

This isn’t about pointing fingers at any one company or institution. The construction company in my last post isn’t the villain — they’re building water treatment infrastructure that communities depend on. The schools aren’t failing — they’re working with the information they have. The workforce programs aren’t broken — they’re doing the best they can with incomplete visibility.

The problem is that nobody has given any of them a tool that shows the full picture. That’s what I’m building.

To educators and program directors: If you could see a real-time dashboard showing exactly which skills employers are demanding that your program doesn’t cover — what would you change first? What’s stopping you from changing it now?

To business owners and hiring managers: Would you use a tool that analyzed your job postings and told you whether you’re asking for one role or three? Would it change how you hire — or how you build your internal systems?

I want both sides of this conversation in the same room. Because the gap between what companies post and what schools prepare students for isn’t going to close itself. Someone has to build the bridge.

I’m building it. Come watch. Better yet — come help.

Nasly Duarte is an AI Solution Architect and accounting strategist based in Miami, FL. She’s currently studying Computer Vision and NLP at Miami Dade College while building tools that bridge the gap between workforce development, education, and the real demands of the job market. Follow her build-in-public journey on Buy Me A Coffee and LinkedIn.

Why Are We Still Cleaning Data in 2026?

 Why Are We Still Cleaning Data in 2026?

By Nasly Duarte 


Data scientists spend 80% of their time cleaning data.

That stat is from 2016. It's 2026 now. A full decade later. And the number hasn't moved.

Think about that.

In ten years we've built large language models that write code, autonomous vehicles that navigate cities, AI systems that diagnose disease from imaging. But we still can't figure out how to stop dirty data from entering a system.

Or maybe we just stopped asking.

The industry's response to the 80% problem has been to build better cleaning tools. Better ETL pipelines. Better data wrangling platforms. Better preprocessing libraries. Faster joins. Smarter imputation. More efficient deduplication.

We got really, really good at mopping the floor.

But nobody turned off the faucet.

Let's look at what "data cleaning" actually means:

Inner joins — combining records from two separate sources that should have been connected from the start. Why are they separate? Because two systems captured the same information independently and nobody enforced a shared structure.

Null handling — filling in missing values. Why are there missing values? Because the system allowed someone to submit incomplete data without flagging it.

Deduplication — removing records that appear more than once. Why do duplicates exist? Because multiple people entered the same information in different places and nothing prevented it.

Type casting — converting text to numbers, strings to dates. Why is a price stored as text? Because a human typed "$6,000" into a free-text field instead of entering a number into a validated field.

Normalization — restructuring data into a consistent format. Why is the format inconsistent? Because the system accepted any format the user felt like giving it.

Every single technique exists for the same reason.

The system that created the data didn't enforce structure at the point of entry.

That's not a data problem. That's a design problem.

And here's what concerns me: we're training an entire generation of data professionals to accept this as normal. Preprocessing isn't taught as a workaround — it's taught as a core competency. As if the mess is a given and our job is just to clean it up faster.

What if it's not a given?

What if instead of building smarter cleaning tools, we designed systems that produce clean, structured, validated, relational data the moment it's created? Not cleaned after the fact. Not transformed downstream. Not wrangled into shape by a team of engineers. Structured from the start.

What would that eliminate?

No more inner joins because the data is already linked at creation. No more null handling because incomplete records are blocked before they enter the system. No more deduplication because data is generated once, from one source, through one process. No more type casting because outputs are typed by design, not by human input. No more ETL because there's nothing to extract, transform, or load. The data is already where it needs to be, in the format it needs to be in.

I'm not saying preprocessing knowledge is useless. If you work with data today, in the real world, you need every one of those skills. The data is messy. The systems are broken. The cleaning has to happen.

But I am saying we should stop treating the mess as permanent.

The 80% number hasn't changed in a decade because we've been optimizing the wrong side of the equation. We keep investing in what happens after data enters a system. Almost nobody is investing in what happens at the point of creation.

That's the question I can't stop thinking about.

What if instead of training people to clean data, we designed systems that never produce dirty data in the first place?

What would that change about how we build technology? What would that change about how we teach data science? What would that change about the 80%?

I'd love to hear from anyone who's working on this — or thinking about it.

#DataScience #AI #DataGovernance #DataEngineering #SystemsThinking #CEFModel

Monday, March 2, 2026

It’s 2026. Why Are We Still Hiring People to Be Entire Departments?

 It’s 2026. Why Are We Still Hiring People to Be Entire Departments?

How one job post exposes an industry-wide systems failure — and what AI could fix tomorrow.

By Nasly Duarte

AI Solution Architect | Accounting & Operations Strategist

The Job Post That Stopped My Scroll

I was scrolling through job postings the other day — something I do regularly, not just for myself, but to study the market. When you’ve spent over a decade working in accounting and operations across construction, retail, and service industries, you start reading job descriptions the way a mechanic listens to an engine. You can hear what’s wrong before anyone tells you.

This particular post caught my eye. Assistant Controller. A company doing wastewater treatment projects across three counties in South Florida. Salary range: $61K to $90K. Benefits included. Sounded reasonable.

Then I read the responsibilities.

General ledger management. Accounts payable. Accounts receivable. Payroll. Monthly, quarterly, and annual financial reporting. Internal controls. GAAP and IFRS compliance. Audit support. Budgeting. Forecasting. Cash flow management. Bank reconciliations. Staff supervision and mentoring. System implementations. Process improvements. M&A support.

I read it again. Then I counted. That’s not one job. That’s three.

Three Roles, One Title, One Salary

Let me break this down, because this is not a matter of opinion. These are distinct, well-defined roles in any properly structured finance department.

Role

Core Responsibilities

Market Salary

Assistant Controller

GL maintenance, month-end close, reconciliations, AP/AR oversight

$60,000 – $75,000

Controller

Financial reporting, internal controls, audit management, staff supervision, compliance

$90,000 – $130,000

CFO

Strategic planning, budgeting & forecasting, cash flow management, system implementations, M&A

$150,000+

That job post asks for all three tiers. At the bottom-tier price. This isn’t a company being intentionally exploitative — it’s a company that doesn’t have the internal structure to know the difference. And that’s a much bigger problem.

The Problem Isn’t the Person. It’s the System They Don’t Have.

Here’s what I’ve learned from working inside companies like this: the bloated job description is never the disease. It’s always the symptom.

When a company posts a role that spans three departments, what they’re really telling you is that they don’t have integrated systems. Estimating lives in one place — maybe a spreadsheet, maybe a standalone tool. Project management lives in another. Accounting lives in Sage or QuickBooks or whatever was set up ten years ago and never revisited.

Nobody reconciles the estimate to actual costs in real time. The project manager knows they’re over budget on materials, but accounting doesn’t see it until month-end close. By then, the damage is done. The variance shows up as a surprise in the financial statements instead of a flag on the dashboard weeks earlier.

So what do they do? They hire a person to be the bridge. One human being to manually connect all the disconnected pieces. They ask that person to reconcile the GL and manage cash flow and build internal controls and run audits and implement new systems and supervise staff. Because without a system, everything falls on a person.

That person burns out. Leaves. And the cycle starts over with a new job post that looks exactly the same.

The AI Elephant in the Room

It’s 2026. Let that sink in for a moment.

NLP models can read and categorize invoices. AI agents can automate recurring journal entries and flag anomalies. Machine learning can forecast cash flow based on historical patterns and project timelines. Automation can handle the repetitive, time-consuming close process that eats up the first two weeks of every month.

Half of what’s in these job descriptions is work that a well-designed system handles — not a person working 60 hours a week trying to hold everything together with spreadsheets and willpower.

The question isn’t whether AI can help. The technology exists. The tools are accessible. The question is: why isn’t leadership asking? Why are we still solving architecture problems with headcount?

I think the answer is simple and uncomfortable: many companies don’t know what they don’t know. They’ve never seen what an integrated system looks like, so they can’t imagine it. They hire executives with titles but no experience in modern systems design. And those executives hire people the same way they were hired — to fill seats, not to build infrastructure.

What This Actually Costs

Let’s talk about the real price tag of this pattern, because it’s not just an HR problem.

Turnover costs. Replacing a mid-level finance employee costs 50–200% of their annual salary when you factor in recruiting, onboarding, lost productivity, and institutional knowledge that walks out the door.

Bad data. When one person is doing the work of three, corners get cut. Reconciliations get rushed. Variances get missed. Financial statements become less reliable, which means decisions are being made on information that’s incomplete or wrong.

Late reporting. When budget variances don’t surface until month-end — or worse, quarter-end — you’re managing projects in the rearview mirror. In construction, where a single project can run into the millions, that delay can be catastrophic.

Human cost. This is the one nobody puts on a spreadsheet. The person in that seat is working nights and weekends. They’re stressed, exhausted, and isolated because no one else in the company understands the full scope of what they carry. They’re not just managing accounts — they’re managing the entire financial nervous system of the business with no support and no system underneath them.

People are being set up to fail by design. Not out of malice, but out of structural ignorance. And that has to change.

What the Fix Actually Looks Like

The good news is this isn’t a mystery. The path from chaos to clarity is well understood. It just requires someone who can see both sides — the accounting reality and the technology architecture.

In a properly integrated system, the estimate flows into job costing. Job costing feeds the general ledger in real time. Reporting is automated. Exceptions and variances surface the moment they happen, not thirty days later. Cash flow projections update dynamically based on project progress and billing schedules.

You don’t need a $500,000 ERP implementation to get there. You need someone who understands the accounting workflows and the technology — someone who can map the process, identify where data breaks down between departments, and build the connective tissue that turns fragmented information into a living, breathing system.

That person exists. Companies just aren’t looking for them because they don’t know to ask. They’re still writing job posts for three people crammed into one title, hoping the right human will somehow compensate for the missing architecture.

Two Questions, Two Audiences

This isn’t about shaming anyone. Companies like the one in that job post are the backbone of infrastructure — they build the systems that give us clean water. They deserve better operational design. And the people they hire deserve to be set up for success, not survival.

So I’m asking two questions, and I want both sides of this conversation in the same room.

To employees: Have you ever been hired for one job and ended up doing three? What did that cost you — not just professionally, but personally? How long did you stay before you realized the role was structurally impossible?

To business owners: What’s really stopping you from investing in the systems that would eliminate the need for these impossible hires? Is it budget? Is it not knowing where to start? Is it that no one has ever shown you what the alternative looks like?

Drop your answers below. I want employees and owners seeing each other’s reality — because the gap between what companies post and what employees experience is a conversation that’s long overdue.

Nasly Duarte is an AI Solution Architect and accounting strategist based in Miami, FL. She builds intelligent systems that bridge finance and operations, and writes about the intersection of technology, workforce development, and human well-being.

Everything You Need to Know About AI Is Already Spinning at the Fair

The Mindful Dollar  ·  AI Explained Simply Everything You Need to Know About AI Is Already Spinning at the Fair You don't need a compute...