Baz Hits $17M Seed, Pushes AI Code Review to Planning Stage

The interesting move from Baz Technologies this week isn't the funding number, it's where the company is trying to put the code review. SiliconANGLE reports that Baz launched a tool called Baz Planner at the AI Engineer World's Fair in San Francisco that runs review against an architecture spec before the code gets written, not after a pull request is opened. At the same time the company added $9 million to its seed, taking the round to $17 million, co-led by Battery Ventures and Boldstart Ventures with AFG Partners and Disruptive VC joining.

The architecture is four agents working in a loop. A spec reviewer checks the planned change against product requirements. An advanced security agent reasons across authorization and network boundaries. A site reliability engineer agent correlates the proposed change against production telemetry. A fixer agent applies the validated change in an isolated runtime. The pitch from co-founder and CEO Guy Eisenkot is the obvious one, that the planning stage is "where bugs and vulnerabilities are cheapest to eliminate." Battery's Barak Schoster framed Baz as a "super harness that coordinates" coding agents, which tells you where the investor thesis sits.

Why this is worth paying attention to if you ship software with copilots: the current generation of agentic coding tools is good at producing diffs and not yet good at producing diffs you can merge without a human or a second model cleaning them up. Moving the review earlier, against an architecture document rather than a finished patch, is a plausible shape for what comes after the "AI writes the code, humans review it" loop everyone is currently stuck in. The company's claim is that early adopters cut downstream rework by more than 65%, measured by reverts and hotfixes.

The honest caveat is that a single-source vendor number from early adopters isn't a controlled study, and "rework reduction" is a metric where definitions do a lot of work. The reporting also doesn't tell you how many customers are in that early-adopter pool, what languages they run, or how Baz handles a planning loop where the spec itself is wrong. Founders coming out of Palo Alto Networks' cloud application security business probably explains the security-and-SRE framing of the agent roster more than the demand pattern does.

If the approach lands, the beneficiaries are the security and SRE teams currently pulled in at the end of every AI-generated change, and the engineering leaders trying to figure out which seat in the agentic stack is worth paying for next.