The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) · Sam Charrington

Agent Swarms and Knowledge Graphs for Autonomous Software Development with Siddhant Pardeshi

·1 hr 16 min·5 clips

Blitzy's Sid explains how they recruit tens of thousands of agents without a single orchestrator bottleneck.

Siddhant Pardeshi joined NVIDIA in January 2016 when it was worth $32 billion — a data point he notes with some irony given current AI valuations. He worked on early generative AI including GANs, autoencoders, and NLP when BERT was still state of the art. After recognizing the potential of transformer-based models, he left for a joint MBA/MS program at Harvard Business School, where he met his co-founder Brian. They founded Blitzy on a bet made when context windows were around 10,000 tokens: that AI would eventually be as good or better than humans at writing code, and that the real opportunity was not code generation but full software engineering. Blitzy's architecture uses database-driven orchestration rather than a single orchestrator agent — the system dynamically recruits multiple swarms of agents, using the database as the orchestration layer. This allows tens of thousands of agents to work in parallel without a single point of failure or coordination bottleneck. Pardeshi explains that Blitzy's system maps the entire codebase, generates an agent action plan subject to human approval, and then executes autonomously — producing hundreds of thousands or millions of lines of code where everything compiles, all tests pass, and UIs are pixel-perfect. He describes reaching 5x developer velocity in enterprise environments. The episode explores what autonomous software development means for enterprises that have accumulated decades of technical debt, and how human engineers shift from writing code to declaring intent and reviewing agent plans.

As heard by us

Autonomous coding is framed as a disciplined, graph-grounded workflow, with a bit of scale gloss around it.

Sid Pardeshi presents a workable system for autonomous software development: multiple agent swarms, sandboxed environments, GitHub commits used as checkpoints, and a graph database that gives the codebase shared structure.

Read the full review in PlayNext →

Why you'd press play

You want a concrete blueprint for scaling autonomous software development beyond one helpful agent.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)