Why CodiumAI Could Pass Four Tests for a High Potential Startup

Back in the dot-com era, the world was awash in new public startup companies — most of which did not survive. In the last year, a new bubble has been expanding — generative artificial intelligence.

The current bubble is very different from the dot-com era, which featured 2,888 initial public offerings between 1996 and 2000. After all, exactly zero generative AI startups have gone public.

However, with the possible exception of ChatGPT creator OpenAI, it’s too early for any generative AI company to go public. For that, companies must generate at least $100 million in revenue and top-line growth of at least 30 percent to 50 percent. On top of that, the IPO market has been inhospitable to tech startups since late 2021.

Based on my experience, a startup has little chance of getting to that point unless it does these four things well:

• Solve a significant source of customer pain
• Avoid competing with established companies
• Field a leadership team with the skill and passion to relieve the customer’s pain well
• Deliver so much value that a fast-growing number of customers will pay for your product

I recently interviewed the CEO of CodiumAI, a Tel Aviv-based startup that passes several of these tests. PitchBook notes that the 19-employee company raised $11 million in seed funding last year.

What it lacks in scale, the company makes up for in talent. In December 2023, CodiumAI launched an open source AI code generation tool, AlphaCodium, “inspired by Google DeepMind’s AlphaCode,” as VentureBeat noted. In January 2024, AlphaCodium surpassed AlphaCode on the most challenging AI code generation benchmarks, “setting X/Twitter all aflutter,” reported VentureBeat.

Read on for how CodiumAI fares on my four tests, and the lessons for your startup.

Solve a significant source of customer pain: Pass

Before you start a company, identify customer pain that rivals are ignoring. For the hundreds of founders I have interviewed, solving their own pain drove them to take the entrepreneurial plunge.

That is true of CodiumAI. “The challenge is code correctness, integrity, best practices, and values,” co-founder and CEO Itamar Friedman told me in a March 14 interview. “This is my life mission.”

“After I got my B.S. degree from the Technion, I worked at [computer network equipment provider] Melanox. It was in the early days of verification [comparing manufactured products to their design specifications]. I asked myself ‘Why don’t we have such tools in software?’ With hardware, it’s bits-in, bits-out. With software, the specifications are soft. This is a problem that AI can [help solve],” he added.

Avoid competing with established companies: Fail

It’s generally a good idea for a startup to avoid competing directly with an established company. That is particularly true when the startup is competing with an incumbent’s core product. The reason is, customers would rather avoid the risk of buying from a company with relatively limited resources.

Since CodiumAI competes with Google and OpenAI, the startup does not fare well on this test. However, since CodiumAI’s focus on code quality and coder productivity is not a core product of either company, perhaps the startup does not fail this test.

Field a team that can relieve customer pain well: Pass

To solve an important problem facing customers, you must field a team with the skills needed to build an industry-leading solution.

By prevailing on that AI code generation test, CodiumAI has demonstrated the skill of its small team. The startup’s success is due to flow engineering. “The reason AI is not generating working code is not because you need a better LLM,” Friedman said. “It’s because you need a flow.”

“Culture will not happen with AI. To make it work, you need to go from prompt engineering to flow engineering. Rather than prompting, flow engineering forces developers to make it more accurate. They must follow a process — define the problem, brainstorm hypotheses, rank the hypotheses, test the best ones, and try new hypotheses in response to test results,” he added.

Win new customers eager to buy your product: Pass

If you can relieve one customer’s pain, you must scale the business by winning new customers eager to buy your product.

CodiumAI is solving this scaling challenge. “We have over 100 paying customers ranging from teams of two to 1,000 developers,” he said. “We are signing more big companies.”

“As an engineer, I am uncomfortable speculating about the future. But as CEO, I try to speculate,” Friedman said. “Software is eating the world and AI is eating software. Companies spend $50 million on developers and boosting their productivity is a $2 trillion problem. I am building a sales team and collaborating with channels in different countries, such as India,” he added.