Agentic AI Could Bring Software’s Profits Forward by Years

2The domino effect

The first domino is engineering productivity. Coding agents, the form of agentic AI built for software development, multiply what each engineer can produce.

With coding agents such as Claude Code, Codex and Cursor, engineers produce more working code in less time, which is why most software developers now use them. Three examples from very different companies show how large and widespread the gain is:[1]

Engineering capacity on that scale can be remarkably inexpensive to add. One 2025 study, known as SWE-Lancer, compared what an AI model cost to run with what developers had actually been paid for the same software work. OpenAI’s o1 model completed $20 of that work for every $1 it cost to run, even after counting the attempts that failed. And o1 is now an older model. Today’s coding agents pair far stronger models with more sophisticated workflows, so we expect the return on that spending to be higher still.[2]

Measured against payroll, the cost is just as modest. Anthropic reports that companies using Claude Code pay $150 to $250 per developer per month on average. That is roughly 1% to 2% of what a software-as-a-service (SaaS) company pays a software developer. For that small addition to payroll, management makes the engineers it already employs far more productive.

Far more output from the same engineers, at little added cost, is more than a change in how software gets written. It is AI’s Trojan horse: it enters a company as a tool for engineers and ends up changing the economics of the business. That is where the opportunity for investors lies.

The economics change because, in a SaaS business, three things move together: engineering headcount, the product roadmap and revenue growth. More engineers build more products, and more products bring in more revenue. Until now, doubling the business has required a large expansion of the engineering team.

Agentic AI is breaking that link. A company no longer needs the same number of engineers to reach the next level of scale, so it can keep growing revenue while adding far fewer engineers than it had planned.

None of this requires layoffs. The company simply hires far fewer engineers than it otherwise would. That avoided spending can be worth billions of dollars in market value, because revenue keeps coming in while less of it goes back out.

For a high-growth software company that barely turns a profit, keeping more of each revenue dollar makes an enormous difference: profit margins expand, earnings grow much faster than revenue, and earnings per share rise sharply.

The pattern itself is standard software economics. A company spends heavily to grow until revenue starts to outpace expenses, and then profits rise quickly. Agentic AI could bring that turning point forward by years, and the market’s expectations do not yet reflect it.

When they do, earnings forecasts will rise, valuations will be reconsidered, and these stocks could be revalued substantially.

Within a few years, this may simply be how software businesses operate, and higher earnings will surprise no one. By then, the largest gains may already have been made. The opportunity lies in owning these businesses before Wall Street catches up.

3A simple thought experiment

A simple example with round numbers makes the effect concrete.

Consider a software company with $1 billion in annual revenue. It spends $300 million on R&D, and after all other costs it keeps $30 million in operating profit. Of every dollar that comes in, three cents is left over.

Next, double the business. At 20% to 25% annual revenue growth, revenue reaches $2 billion in roughly three to four years.

Software businesses already benefit from operating leverage, meaning their costs tend to grow more slowly than their revenue. We therefore assume R&D spending grows 85% while revenue doubles, and we hold all other costs at the same share of revenue. R&D reaches $555 million and operating profit reaches $105 million. That is a good result: revenue doubles and operating profit more than triples.

Now introduce coding agents. Suppose management delivers the same revenue growth while increasing R&D spending by 50% instead of 85%.

The business is still investing heavily in product development: R&D spending rises by $150 million. Management simply avoids the additional $105 million it would have spent on the baseline path.

That $105 million becomes operating profit, on the same revenue and the same growth. Revenue doubles, and operating profit rises sevenfold. Exhibit 2 lets you set the R&D growth rate yourself.

4The domino effect continues

The step-change in R&D efficiency gives shareholders two further advantages: faster revenue growth and a higher ceiling on profit margins.

Coding agents do more than help software companies earn more on the revenue they have. They can also help them grow faster.

A large part of revenue growth depends on how much a business can spend on sales and marketing, because even a strong product needs sales and marketing to reach customers. Resources are limited, and money spent developing products cannot be spent selling them. Management therefore has to divide its budget between sales and development.

Coding agents change that trade-off. As engineering becomes more efficient, management can redirect some of the savings toward expanding the sales force and acquiring more customers, which sets off a second domino effect.

More salespeople bring in more customers, and more customers accelerate revenue growth. That additional revenue then flows into a business with better profit margins, so faster revenue growth and higher margins combine into even faster earnings growth.

Shareholders benefit twice. They earn more on the growth the business would have had anyway, and they may also own a business whose growth accelerates. Earnings grow larger and arrive sooner.

5What it could mean for the share price

In the end, what matters to an investor is how much the share price rises.

We think about valuation as a company’s earnings multiplied by the price the market is likely to pay for each dollar of them, known as the price-to-earnings (P/E) multiple. The aim is direction more than precision: an estimate of a sensible share price under reasonable conditions. We prefer a defensible fair value to a “precise” valuation every time.

In the doubling example with coding agents, operating profit rises from $30 million to $210 million. Assume no net interest expense, a constant 25% tax rate and 100 million diluted shares. Earnings per share rise from about $0.23 to $1.58.

Start the stock at 100 times its current earnings, which puts the initial share price at $22.50. A P/E of 100x assumes the market already expects substantial earnings growth and operating leverage.

As that profit arrives, the multiple often compresses materially. Our base case cuts the P/E in half, from 100x to 50x, over the same period in which revenue doubles.

Earnings per share grow sevenfold while the multiple halves, so the modeled share price reaches $78.75. A hypothetical $100,000 position becomes $350,000.

A dramatically better earnings trajectory is also likely to attract investor enthusiasm. If the P/E keeps three-quarters of its starting level and ends at 75x, the position becomes $525,000. If enthusiasm holds the multiple at 100x, it becomes $700,000.

6A shortlist for further research

There are hundreds of publicly traded SaaS companies, and even more whose core services run on software. That is a lot of stocks to examine.

We have done some initial filtering so that readers can move straight to deeper research. The five companies in Exhibit 7 share the characteristics that make them worth examining through the lens developed here: heavy R&D spending relative to revenue and modest or negative operating margins.

We are not recommending the purchase of any of these stocks. It is a list of candidates for further research. We do believe that some of the biggest beneficiaries of agentic AI are among them.

7Why “SaaS is dead” is the wrong conclusion

One last point concerns the narrative that “SaaS is dead.” We do not accept it. A successful SaaS business is much more than its code. If code were the only moat, these companies would not spend 50% more on sales and marketing than they do on R&D.

That selling effort also teaches them what customers actually want, an insight a new competitor cannot simply copy.

They have also spent years building specialized products, integrating themselves into their customers’ operations, and earning the trust of companies that depend on their software for mission-critical work. An AI startup will not displace them simply because writing code has become cheaper.

We see little sign that the frontier AI labs want to become specialized SaaS businesses. Nor are customers better off building and maintaining the software themselves, because they lack the domain specialization that makes agentic AI work in practice.

Customers also have easier savings available. Using agentic AI to augment their own workforces would save them far more than cutting SaaS spending, and companies are already moving in that direction this year (Exhibit 8). [4]