Hi {{first_name}},

Last week I found a wire confirmation tucked in a drawer. It was from 2015, when my father and I pooled our money and wired $475,000 into a coding academy in Los Angeles. It was the first serious check I'd ever written for a business I wasn't personally running.

The company was Codesmith, founded by Will Sentance and Alex Zai. I'd only known them a few months, but I believed in Will before I even understood what he was teaching. Front end, back end, every coding language, he knew it all.

Around the same time, I was writing what became Buy Then Build. I'd be up before dawn most mornings, sitting in my living room in a Codesmith hoodie, working through ideas that could change how people built wealth. Will and I talked often about what was happening across the industry. He was using the phrase "machine learning" years before most people I knew had even heard it. They thought far ahead, building curriculum for jobs that didn't exist yet.

For years, it went brilliantly. Codesmith grew from $2.4 million in revenue in 2017 to $18.5 million in 2022. Then ChatGPT launched. For a few months, nothing much changed. Then AI started doing some of the work junior developers used to do, and entry-level developer jobs disappeared at a rapid rate. Coding academies started closing. Codesmith's revenue eventually fell to $5 million. I watched a business I believed in get run over by the very technology its founders had been building toward.

Then their long game began to pay off. The constraint in AI has moved from compute to deployment, and the hardest work now is getting AI running inside organizations. Codesmith had been preparing for this already. Today, government agencies from the U.S. Treasury to New York City want people who can put AI to work, and the curriculum Codesmith built years ahead of time is what they needed. That wire confirmation now reminds me of what I saw back in 2015.

This week, I want to talk about how the constraint in AI has already moved once, and why most of the money chasing AI is still aimed at where it used to be. Inside this issue:

— Walker Deibel
WSJ & USA Today Bestselling Author of Buy Then Build
Founder, Build Wealth

P.S. We're now investing in Codesmith through BuildForce, our new SPV. Read on to see why, then find the full deal drop in our portal. 

Please see disclaimers at the bottom of this email and in the presentation. This email is for educational purposes only, shares personal investing decisions that I’ve made and why, and includes forward-looking statements. Potential investors should always conduct their own diligence and read the full PPM.

AI has its own vocabulary. Before digging into what's driving private investment and where it's getting stuck, here are a few terms to understand.

If you are planning to add AI to your portfolio, you've got plenty of company. In S&P Global's LP survey, about 74% of respondents plan to commit capital to AI. That's far ahead of healthcare and longevity, the next most popular investing categories, which sit at roughly 50%. The most eager investors include multi-family offices where close to 90% plan to invest in AI, and wealth managers, about 85%.

According to Goldman Sachs, the five largest U.S. hyperscalers, Amazon, Alphabet (Google), Microsoft, Meta, and Oracle, plan to spend roughly $800 billion on AI infrastructure this year, up 94% from 2025. The same research projects that cumulative AI spending on compute, data centers and power will reach $7.6 trillion between now and 2031. AI firms captured 61% of all global venture capital in 2025, and IT infrastructure and hosting pulled in $109.3 billion of that, more than any other AI category, according to the OECD.

Nearly all of it is meant to buy capacity with more chips, more data centers, more power. The bet is that once the capacity exists, the implementation will take care of itself. This is where things are getting stuck inside companies who want to use AI to better their business. 

The Weakest Link

Picture AI as a four-link chain. It all has to work together. 

When you put resources, whether that's time or money, against the links, you make strong links stronger, but the chain's output can only move if the weakest link is moving. In 1984 Eli Goldratt wrote The Goal to influence businesses to think about continuous improvement in everything they did. The book details how each chain will have a weak link. 

So which link is weak in AI investment? Start with the model, because it's improving faster than anything else in the chain. Stanford's 2026 AI Index tracks how well AI handles work like coding. Two things stood out.

So, the model is holding up its end of the chain.

Next is compute. The spend across the AI industry speaks volumes. As mentioned earlier, the five hyperscalers will spend roughly $800 billion this year, and $7.6 trillion in infrastructure through 2031. The chain has plenty of chips and data centers in the works.

That leaves deployment. This is where the chain breaks. MIT's Project NANDA found that 95% of generative AI pilots produce no measurable financial return. The models work and the compute is there, but most companies aren’t getting the tools themselves into production. The investment dollars aren’t being pushed into deploying the work itself.

At the end of the chain is value. McKinsey estimates generative AI could add $2.6 trillion to $4.4 trillion in economic value every year once it's fully applied across industries. For perspective, that's close to the GDP of the United Kingdom, and all of it is waiting on the weakest link to get AI working inside businesses.  This forms a bottleneck where plenty of compute is ready to be used but it can’t get through deployment fast enough. 

Understanding the Deployment Constraint

Very little money is aimed at deployment. Corporate training, the market for retraining workers, is a $391.1 billion industry. The slice built for AI-specific transformation is projected to reach $125 billion by 2030, according to Training Industry. The five hyperscalers will spend more than six times that on infrastructure this year alone. The answer is in staffing up the teams that can push this deployment within companies. 

Employers are feeling the pinch first. Job postings for forward deployed engineers (FDEs) rose more than 1,000% between January and August compared with the same period last year. These engineers will be the ones that can move an AI model or tool out of a test environment and integrate it into daily business operations so employees or customers can actually use it. Reducing the bottleneck that is growing around deployment. 

Before You Buy In

Getting AI exposure is easy. Newly funded AI companies jumped from 2,050 in 2024 to 3,499 in 2025. The harder part is knowing which link you're buying. The big money is going to the models and compute. However, the Deployment Constraint is where the capital is needed. 

Want the full map? Our report follows private capital across the AI stack to see where investors are going, what gets the most funding, which states are winning (i.e., California alone has pulled in roughly $218 billion), and how fast new AI companies are forming.

How Private Investors are Investing in AI

Codesmith is one company that has worked inside the Deployment Constraint for years, well before anyone was tracking it.

Knocked Around by AI, Now Hired to Deploy It

Will Sentance and Alex Zai started Codesmith in Los Angeles in 2015 as a coding academy. From the start, they measured success by the percentage of graduates placed within ninety days of finishing the program, and the average starting salary of each cohort. By its own tracking, Codesmith has ranked first on both in nearly every cohort since that launch.

In 2022, the market started to move.

The Hit

Codesmith brought in $18.5 million in 2022, the same year ChatGPT launched. And through 2023, revenue held steady at $18.3 million.

Then companies started hiring far fewer entry-level software engineers, the job Codesmith trained people for. The drop came faster than a lot of coding schools or their students anticipated.

A 2025 Stanford study using ADP payroll data found that employment for software developers aged 22 to 25 fell nearly 20% from its late-2022 peak. In our own deal presentation, we put the toll at roughly one in ten coding academies nationally going out of business over that stretch.

Codesmith's revenue fell to $9.6 million in 2024 and $5 million in 2025.

The Pivot

The same technology that shrank Codesmith's market was also creating a new job need. Thankfully, Will and Alex had been building the curriculum for it since 2017. They knew the deployment needs would eventually catch up. 

The idea was to take people who already understand a business's domain, and give them the technical depth to deploy AI into that business themselves. For years, the role itself didn't have a name. Eventually it was labeled forward deployed engineer (FDE).

The Pipeline

Once the role was identified, organizations started looking for people who could fill it, and that changed who Codesmith was targeting. Now its customers are government agencies and large companies that need forward deployed engineers. Codesmith trains people to do the work and places them inside the organizations that need them. These relationships include:

That last one shows the value of this pivot. When a client pays $300,000 for one forward deployed engineer placement, that fee goes to Codesmith. Fill half of that 200-engineer plan, roughly 100 placements, and that's $30 million in revenue before the contract is fully staffed. That's six times what the company brought in last year.

Microsoft and EY have committed more than $1 billion over five years to a joint initiative that pairs Microsoft's forward deployed engineers with EY's industry teams. A Microsoft executive working directly on that initiative evaluated Codesmith's approach. Her read was that the market has plenty of capable coders and nowhere near enough people who can turn that capability into a business outcome, and that Codesmith has been building for that gap longer than most.

A Deloitte human capital transformation executive, previously a senior federal workforce official at the Commerce Department, advises the company and holds equity in it. Having that knowledge inside has been vital to Codesmith. This executive has seen workforce problems from inside a major consulting firm and inside the federal government, the two markets Codesmith now sells to.

Firms like Accenture and Deloitte can win AI contracts on brand awareness and existing client relationships. Then they have to staff the contracts they win, and Codesmith supplies the people. So the problem has flipped. Codesmith now has more demand, and hiring needs to keep up.

What to Keep an Eye On

None of these plans come without some risk. Codesmith runs with fourteen full-time and part-time people. Being able to scale to meet a 200-engineer hiring plan or a multi-agency contract is going to require significant legwork. Government contracts pay on the government's schedule, and fulfilling them takes people up front. That's part of why Codesmith is raising capital now.

The client pipeline is real and robust, but most of it isn't fully closed. Today it still rests on a small number of relationships instead of a broad client base. More of Codesmith’s forward-thinking innovation will be required to drive the next phase.

Every company trying to make money from AI runs into the Deployment Constraint. It makes it hard to find a company that understands how AI works inside a business. That’s when you run the Constraint Test, a four-question test to help you identify a good AI deal before you commit. To show how it works, we'll run Codesmith through each one.

1. How much of the growth story is signed, and how much is a ceiling?

Federal purchase agreements, enterprise pipelines and letters of intent all point to future revenue that hasn't been booked yet. Codesmith's Treasury blanket purchase agreement (BPA) is a good example. It sets a ceiling of up to $118 million (the most that could be spent under it), and that amount is shared among a small number of awardees. Using only the numbers in the pitch, you should be able to tell how much revenue is already signed and how much is still only possible.

Look closer if the pitch blurs the two, or you can't tell which numbers are signed.

2. Where did the operator's pipeline come from?

A pipeline is the list of deals a company expects to close, and where those deals came from tells you a lot. Codesmith's federal and enterprise deals started with its own alumni, people already working inside the agencies and companies now signing contracts. Picture an engineer who went through Codesmith years ago. When their employer starts looking for help with AI, that engineer recognizes the training and vouches for it before anyone from Codesmith makes a formal pitch. That's a different kind of asset from a pipeline built through cold calls and sales meetings, and it shows up in how fast deals close and how many there are.

Look closer if you are asking who inside the client opened the door and the company can't give you a name.

3. Who's the likely buyer, and why can't they build this themselves?

This question is about the company's exit: who could realistically buy it someday, and could that buyer build the same thing on its own? Start with where the company fits in its market. The largest consulting firms win these AI contracts on brand and existing client relationships. Codesmith's advantage is its bench: it can staff what the large firms win, on the timeline their clients expect. That staffing capacity is what makes a services business worth acquiring.

Look closer if the likely buyers could build the same capability themselves in under a year. If they can, the company doesn't have a real moat, meaning an advantage competitors can't easily copy.

4. Can the team deliver the growth it's forecasting?

A growth forecast is only as good as the team behind it. That's especially true in a services business, where revenue grows only as fast as the company can hire, train and put people to work. Strong demand doesn't help if there's no one to answer your call.

That Microsoft executive who evaluated Codesmith's approach said she’d want to know, “Can a small team staff fast enough to keep up with demand?” It's a fair question.  A fourteen-person team pursuing federal, enterprise, and municipal contracts at the same time succeeds or fails based on its ability to hire fast enough.

Look closer if the forecast depends on more hires than the company has shown it can handle.

"AI deployment" is an easy label to claim and a hard one to prove. It takes signed contracts, trusted relationships, skills that are hard to copy and a team that can hire fast enough to keep up. Most companies using the label are missing at least one. Run these questions on the next pitch that calls itself "AI deployment." When you find the rare diamond in the rough, it's because you put in the work.

I wrote my first check into Codesmith in 2015, and I'm still an investor today. Eleven years later, the job it trains people for finally has a name, forward deployed engineer, and organizations are signing on to fill it.

That opportunity is why we created BuildForce, our new SPV investing in Codesmith. BuildForce is joining Codesmith's new round through a SAFE (simple agreement for future equity). The capital will help Codesmith staff up to fulfill its Treasury purchase agreement and give it the budget to close more of its pipeline.

Please see disclaimers at the bottom of this email and in the presentation. This email is for educational purposes only, shares personal investing decisions that I’ve made and why, and includes forward-looking statements. Potential investors should always conduct their own diligence and read the full PPM.

"The bottleneck is, after all, always at the head of the bottle." - Peter Drucker

Every day, a line forms in St. Peter's Square. Thousands of people wait near Bernini's colonnade for their turn at the entrance. They know the value of this experience, but still they have to wait at the door.

Drucker's point was that an organization can only move as fast as its leaders let it. AI is stuck at that door. Leaders can approve the budget and buy the tools, but nothing changes until someone inside the company knows how to put those tools to work. Widen the door, and trillions of dollars in value could start to move.

Already a reader? Help other investors open their eyes to the private markets and forward this to someone who would value the insight. Or hit reply to tell us what you think.

By continuing to receive the Wealth Stack newsletter, you acknowledge and accept our Privacy Policy, which includes the sharing of subscriber engagement data with select advertising partners.  

To unsubscribe, click here.

This is not financial advice. Illustrative output of a reasoned thought experiment. Not a backtest, guarantee, or prospectus. Actual results vary based on market conditions, fund selection, timing, fees, taxes, and factors not modeled. Private credit, CRE, and leveraged strategies involve significant risk including loss of principal. Consult a qualified financial advisor.