Programming · News · 🇧🇷 Brazil
Venturus builds AI agents for software development and product planning drops from 8 hours to 1
The Campinas innovation center measured internal product, testing, coding and design tasks, and total time fell from 121 to 45 hours.
- Profession
- Developer, Product Manager, Designer
- Tools
- VNT Gen.AI
- Source
- tv7.com.br ↗
- Country
- 🇧🇷 Brazil
- Report date
- November 3, 2026
- Verified on
- November 4, 2026
See more: AI for developers · AI for product managers · AI for designers
What they did
Venturus, a technology innovation center, started building VNT Gen.AI in August 2023, a platform for creating custom AI agents for each stage of software development: writing requirements, planning and estimating, building and running tests, reviewing code and prototyping screens. The first internal version came out in the first half of 2024, and today the project has 14 people. The same technology also looks up company information and analyzes documents; in one example from the power sector, an agent reads substation documents and builds a checklist for approval, with a person validating in the flow.
How it went
In Venturus's internal assessment, time for the set of tasks fell from 121 to 45 hours, or 63%. In product, specifying, planning and estimating went from 8 hours to 1. In quality, planning and running tests fell from 29 to 6 hours. For developers, from 76 to 35 hours, and for screen prototyping, from 8 to 3. The company says specifications with context, business rules and well defined scenarios are what helps the agents most.
Good to know
The measurement was done by the organization itself, which also sells the tool, and the article notes that the numbers don't show savings in money or guarantee the same result at other companies. The gain varied a lot from one area to another.
The prompt
I'm a product manager at [company] and I need to specify the feature [describe in a few lines] for the [technology] team. Write the spec with the goal, business rules, usage scenarios (including error cases), acceptance criteria in given, when, then format, and the questions I still need to answer. Then break it into front end, back end and testing tasks with an effort estimate in hours, flagging the estimates you're least sure about.
Replace what is in [brackets]. Prompt written by us from the account, not a literal copy of the source. Do not paste other people’s personal data (privacy laws such as LGPD and GDPR apply).
Summary in our own words. Results as stated in the original account; we have not tested or audited them. Read the source ↗ Found a mistake?
Similar cases
01Team turns on the strictest compiler settings on a 1.5 million line system and uses Claude to fix what comes up
A developer explained on Hacker News how Claude tackles whole classes of compiler errors, with the team's own skills and scripts, in a huge .NET codebase.
02A requirements doc shelved for a year becomes a live website in one day with Claude Code
A tech consultant pulled out the spec for an aggregator of his own posts and shipped the site in a single day with Claude Code.
03Livelo engineers report up to 50% higher productivity with AI agents
The Brazilian loyalty program brought Kiro and Amazon Q Developer to its team of about 240 engineers, with metrics and governance from day one.
04A leaner MCP server finds wasted AWS spend in fewer calls
A cloud architect rebuilt a 15-tool MCP server with just 4 tools, and the agent now answers where an AWS account is wasting money in 2 calls instead of 7.
05Letting the agent deploy and read production errors on its own
The founder of a deployment platform explains how they designed an MCP server to take the human out of "copy and paste the log".
06Launching a habit app solo and doing the marketing with AI
A dev created Kada, a habit app with no account and no ads, and used AI for marketing.
07Building an "internet TV" in 6 days with a team of agents
A dev created Crom TV, with 24-hour channels of YouTube videos, coordinating several AI agents.
08Asking AI for a specific algorithm for a research project
A developer skeptical of LLMs used Gemini to generate random quantum circuits in the QCOP project.