Agentic AI is on the rise — some reports predict that 25% of businesses will deploy it next year (whether they're ready to or not). AI company Blitzy believes it has the autonomous powerhouse to keep up, without sacrificing human oversight.
On Wednesday, the Massachusetts-based company announced Blitzy Platform, an agentic system that "batch-builds up to 300,000 lines of code in a single 8-hour inference run, cutting end-to-end development timelines from months to days." The platform interprets product roadmaps, generates documentation, and identifies final steps for human developers, writing anywhere from 50% to 100% of the necessary code, depending on the project's complexity.
Also: OpenAI releases a slew of developer features in a 'Mini Dev Day'
According to the release, Blitzy Platform works "by orchestrating multiple LLMs to collaborate to achieve 8 to 12 hours of 'thinking time,' during which thousands of specialized, purpose-built AI agents cooperate, plan, build, and validate software based on technical specs." That timeframe is how long it takes for Blitzy's agents to build out a full project, which can be up to 300,000 lines of code for specific enterprise products.
In an exclusive interview and demo, Blitzy walked ZDNET through the platform, explaining that it can generate entire repositories — "hundreds of thousands of lines of code where a copilot might give you a hundred or a few hundred lines of code," as CEO Brian Elliott put it.
Blitzy aims to eliminate the learning curves developers have to scale to get the best output from a copilot. "We've taken the prompt engineering and done it on our end, so that developers work with things they do on a daily basis, like, for example, documentation, PRDs, and technical designs," said CTO Sid Pardeshi during the demo.
Author Daniel Kahneman thought of cognitive capabilities as being in two categories, which can be applied to AI: System 1 and System 2.
Also: Amazon's new Nova AI models could be ground-breaking – why we can't know for certain
While System 1 focuses on speed in its results, such as a chatbot running a model with very low latency that can answer your question in seconds, System 2 takes its time to complete more complex tasks using deeper reasoning. Blitzy aims to lead in System 2, optimizing any model a developer prefers for reasoning.
"We've unlocked the ability to make any model on the market — be that Claude, Llama, any other model that comes out — we can make that think like [OpenAI's] o1, but not just for seconds," said Pardeshi.
He stated that Blitzy's strength lies in its ability to "estimate the complexity of a software project and to recruit the right amount of agents to solve that problem," based on each model's context window.
Also: OpenAI's o1 lies more than any major AI model. Why that matters
However, ZDNET's David Gewirtz — a former computer science professor who has coded commercial products with hundreds of thousands of lines of code — had a different perspective. "While I haven't had a chance to test it and it seems like Blitzy does have something here, it's far from a no-programming solution," he shared. "I think the best way to think of it is as a power tool for some of the mid-stages of traditional enterprise data processing projects."
Gewirtz continued: "In the right hands, it could save businesses a lot of time to deployment. But they'll need very expert coders and designers to get the most out of this tool."
Also: Programmers, do you get window-switch whiplash? Google's new Code Assist aims to help
Blitzy Platform also tries to reduce the error correction developers must do based on their language of choice. "We've manually added support for each language — for example, we've built a compiler in-house that compiles Typescript that addresses these nuances with the package manager," Pardeshi explained. "By the time the customers receive the code, it is ready to run."
Compiler support for Typescript and Python will be available later this month. The company will eventually add support for Java, PHP, C#, C++, Android, and iOS.
How to try it
After booking an onboarding session, you can try Blitzy Platform for free, with limitations, by signing up with your work email. The company offers a seven-day free trial of its Pro tier, which is then $1,000 per month and includes roughly 600,000 lines of code and weekly feature updates. Enterprise clients can "join limited cohorts as Blitzy scales its inference compute capacity" by contacting sales.
We'll be testing Blitzy in-house soon — stay tuned for a full review.