Acceleration of AI Progress Shocks Almost Everyone, Including the Software Engineers Who Build AI

The cadence of technical progress in AI has compressed, and even seasoned engineers are finding it hard to keep up.

In our world of work, one trend stands out: how fast AI changes writing code. GitHub says 92% of developers already use AI assistants. In fact, AI systems now generate more than 35% of new code at Microsoft. We even gathered an “emergency” meeting of our portfolio companies in March because startup founders are also worried about falling behind.

Replit, a “vibe coding” platform where anyone can type in plain language the software they want to see, now reaches 30 million users. Google’s Sundar Pichai says Replit gives developers power that they haven’t seen in 25 years, and Franklin Templeton’s Jenny Johnson says creatives will now have the ability to bring all their ideas to life.

One wrinkle: all that code AI generates may come at a cost. Does it create new “tech debt”? Does AI-generated code work in real-world applications? As companies automate more work once handled by junior engineers, what will hiring new grads look like?

New AI models come out on a near-monthly cadence, and each one continues to impress more than the last. OpenAI just released GPT-5, which immediately replaced all their earlier models.

Startups like ElevenLabs, Cursor, and Synthesia have made rapid progress across voice, code, and video. xAI’s Grok 4 initially stunned many with its reasoning capabilities (especially having only started two years ago), and then sparked backlash for using Elon Musk’s public stances to generate antisemitic responses. In China, open source model Kimi K2, developed by a startup backed by Alibaba, beat OpenAI and Anthropic models on some coding benchmarks at lower cost, and Zhipu released a competitive model to DeepSeek, which continues to quietly improve its logic and reasoning.

Some believe this will accelerate dramatically, pondering catastrophic scenarios.

As AI becomes a core part of how work gets done, institutions beyond tech are starting to prepare people: Roy Bahat, head of Bloomberg Beta, proposed a new national center for training teachers in AI, as a partnership between Microsoft, OpenAI, Anthropic, and the American Federation of Teachers (AFT). We hope to get insights on how AI can improve work for educators, and use it as a model for many other occupations.

At the same time, a quieter narrative has started to emerge: AI might just be… normal. Like electricity, where impact comes through adoption and diffusion, the effects (like brain rot or changing our conception of what it means to be a person) of AI could take time (maybe decades!) to arrive. And new organizations are springing up to sift through the noise, including the Golden Gate Institute, created by a Bloomberg Beta-backed founder (whose first startup created Google Docs).

Despite these questions about AI’s future, AI’s present certainly consists of hefty, growing investment. In 2024 alone, AI-focused companies raised ~$110 billion, up 62% from the year before, though this is concentrated in just a few companies. Ahead of their IPO, CoreWeave acquired Weights & Biases for $1.7 billion to streamline AI model development. And separately, companies are pouring hundreds of billions into building AI data centers to meet rising compute demands, meaningfully affecting GDP.

Meanwhile, with traditional M&A still tight due in part to antitrust scrutiny, AI companies have leaned into alternative acqui‑hire strategies. Google’s hiring of Windsurf’s CEO and key executives, followed by the startup’s absorption into Cognition, echoed a model first pioneered at scale by Microsoft’s “acquisition” of Inflection’s team. Similar moves by Google and Character AI and Meta with ScaleAI (and more recently, its recent AI research hiring spree), demonstrate the battle for AI talent. Still, these structures can be a bad thing (for employees) and can leave deserving teammates behind.

Same fund, same team, and a few victories.

We just celebrated Bloomberg Beta’s 12th birthday, and started investing out of our fifth fund to keep doing what we’ve always done: bet early on the future of work and AI.

Our new fund is the same as every prior one, including the same three equal partners, Roy Bahat, Karin Klein, and James Cham. We’re grateful to Bloomberg LP for the steadfast support.

And since we created Bloomberg Beta:

●  Founders we’ve backed have gone on to raise more than $10 billion.

●  10 startups we’ve backed are worth more than $1 billion (Flexport, Replit, MasterClass, and more!), including Shield AI, whose most recent round valued them at $5.3 billion.

●  Startups we back outperform the market, worth $24 million more at their next round than the average startup

And we’re proud: Business Insider again named Karin Klein one of the top 40 women early-stage investors.

We welcome founders you send our way, especially those working on “hot swaps” where they compete with a successful incumbent using AI.

— The Bloomberg Beta Team

Bloomberg Beta, the early-stage venture firm backed by Bloomberg LP, invests in startups making work better. It was the first venture fund to focus on investing in the future of work, and in artificial intelligence.

Note: Of the companies named above, Bloomberg Beta is a shareholder in Weights & Biases, Replit, Shield AI, Flexport, MasterClass, and xAI.