TikTok's Ex-Algorithm Chief Launches Verdent AI Coding Tool | Hacker News Reader

TikTok's Ex-Algorithm Chief Launches Verdent AI Coding Tool

thenewstack.io

7 pointsnkko8 comments

I have been using GitHub Copilot and other VS Code agent extensions like Cline. Common gaps across these vendors I identified: Planning rarely includes clear acceptance checks, edge-case lists, or traceability from plan items to commits/tests.

I think Verdent provides good planning, where prompts can be structured and I can review, accept, or adjust this plan before any code changes are made.

The only thing I do not like about Verdent does not offer a free plan. It's worth it as a paid tool but it would be great to have free credits to get started. For example, you can use the Cline or Copilot free version or you can bring your own model.

Thansk for the comment. Planning and verifiability is crucial for any serious coding task. Through our early access program we have been providing free credits and we have used that insight to structure the pricing plan. ATM we are not offering free plan as we are focusing all of our efforts to create the best possible experience for engineers.

Thanks for explaining this! I just think of a new developer experience that wants to try out Verdent and pay after. I usually do not pay for a tool that I have not tried and see what it does. Showcasing content or demos is not enough. Take the example of Lovable, they offer 5 credits to send few prompts and if you want to continue, you need to pay.

Thanks for the suggestion, we will take it into account while we are figuring out the next steps.

Thanks for sharing. I’m Zhijie Chen, founder of Verdent and formerly Head of Algorithms at TikTok. Verdent is built for larger, production-grade tasks: it plans before coding, runs agents in parallel, tests results, and explains diffs. Our north star is a full Plan Code Verify loop, and we are building toward that. Curious what kinds of tasks you would actually trust an AI to handle end to end.

Hi Zhijie Chen,

Do you have a benchmark of how many tokens Verdent agents can save for complex asks? For example, I stopped using Cline because it consumes quite a lot of tokens for bigger projects just to analyze and provide context.

Hi, atm we can say that it's moderately less than Cline but we need more testing to give exact numbers. We have plans for testing by users and 3rd parties in future, for full transparency.

I had access to Verdent during early access over the past few weeks, and I've found it to be better than my previous experience with Roo Code + Gemini/Claude.

Reply on news.ycombinator.com