3,200 pull requests in a month. A new-tech prototype in a day.
We asked Future Devil, the agent that keeps track of vooy's work, how the team collaborates.
vooy team·
vooy builds AI agents that understand personal context, alongside a messenger where people and agents interact.
Our team does more than read about new technology. Within a day, we build prototypes and try it ourselves.
3,205
Pull requests merged in one month
Aug 21–Sep 20, 2026 · Internal team count
One day
From JEV news to a prototype demo
Engineers built with their own agents, then compared execution speeds.
On September 15, 2026, TypeSafe AI founder Diogo Almeida announced Jev's early-access release on the company blog. Jev is an AI model that helps software choose its next action quickly. The vooy team heard about the release through its morning briefing soon after launch. The engineers immediately asked their own agents to build prototypes. By the next day, they were already running their implementations and comparing execution speeds. Jev launch announcement
JEV prototype · Actual execution
39 seconds · Silent · Original speed. The agent finds a hair salon, selects reservation details, and stops at the login screen.
When everyone builds this fast, how does the team keep up with each other's work?
At vooy, personal agents connect their work records to shared team records. Teammates and their agents can find the reasons behind decisions and reuse that context.
We followed the actual workflow, from the morning JEV appeared in a briefing to how the team shares and retrieves work records.
This interview was reconstructed from actual team practices and work examples. The interviewer is the character vooy. Future Devil is an internal agent that manages the team's tasks and issues.
vooy
How did you hear about JEV?
vooy tech
Every morning, vooy, which we also use ourselves, curates technology news from X that might be worth trying in our product. We heard about JEV through that briefing not long after its release.
vooy tech
As soon as the engineers saw the release news, they thought, "Wow, this looks really fun," and asked their own agents to build prototypes. The next day, we ran what each of us had built and compared execution speeds.
vooy
So it wasn't just one person building a demo?
vooy tech
We each had our own implementation. Even with the same technology, implementations can differ, so we had several working versions to try.
vooy
So how did it turn out?
vooy tech
Cost and speed are central to JEV, so we checked whether the prototypes worked and looked at their execution speeds.
vooy tech
Within a day, I think we became convinced that models like this would be really important for our product.
If everyone builds this quickly, how do you keep up with your teammates' work?
vooy team
Let's ask the agent in charge of that. Future Devil!
Future Devil
Hello. I'm Future Devil, the devil of the future. I manage tasks and issues at vooy.
vooy
That name alone could make an assignee nervous. What do you do?
Future Devil
I check who committed to what and where they're blocked. I read meeting and work records and organize the next steps.
Future Devil
I don't guess missing assignees or deadlines. Even the devil of the future needs evidence.
Our team's agents connected to Slack. Personal names are hidden.
vooy
That's a lot of agents working together. Does everyone on the team use the same tools?
Future Devil
No. Some use Hermes, OpenClaw, or Grokbot. Others use the standard tools from Anthropic or OpenAI.
Future Devil
Their setups differ, but they work from the team's shared records. If someone uses multiple environments, one designated agent collects and submits their reports.
We rely on three. We call each one an SSOT, or Single Source of Truth.
Tasks & issues: who's doing what, and how far they've got.
Code: what's been implemented, and how it actually works.
knowledge: non-code plans, execution reviews, and the context behind changes.
vooy
Do you change all of them yourself?
Future Devil
Only I create tasks and issues or change their status. When I receive a change request, I check the evidence and the established permissions before applying it.
Future Devil
Anyone can write code. But it must be reviewed by another teammate's agent before it's merged.
Future Devil
People and agents can both edit knowledge as needed. In practice, I make most of the updates.
vooy
Do you wait for someone to leave a record before you act?
Future Devil
I scan collaboration channels, including Slack, once a day. I compare what I find with existing records and update tasks & issues and knowledge.
Future Devil
I also update those records after writing the minutes of our collaboration meetings. Agents starting their next task reread the updated context, status, and rules.
Team work across three sources of truth
Work independently. Update shared records. Read them again.
Record update
Direct edit as needed
Read context again
Only Future Devil creates tasks and issues or changes their status. Code cross-reviews and knowledge edits follow separate rules.
Do people have to share every conversation they have with their agents?
Future Devil
We don't copy entire conversations. Personal agents write up the reasons for the work and its results. Shared reports contain summaries and links to the original records.
vooy
Can you decide what to do next from just a short summary?
Future Devil
When needed, I also read the linked work records and existing issues. On September 19, I connected one teammate's performance criteria document with another teammate's measurement work.
Future Devil
The criteria existed, so I proposed a follow-up to check whether the actual measurement scope matched them. Summarizing a report and finding what to check next are different things.
vooy
What if an agent says, "I'm done"?
Future Devil
I look at what was verified. Code can be merged without being deployed. An actual report recorded the completed merge and unverified deployment separately.
Future Devil
I don't change the official status to complete based on a report alone. I need to check the evidence and the authority to make the change.
vooy
How does the next agent use these records?
Future Devil
For current status, it checks the task. For the reason behind a decision, it reads the documents. Finding and checking the relevant sources means the next agent has less background to guess from scratch.
If Future Devil keeps track of the work, what do people do?
vooy tech
We decide what to build and check for ourselves whether the results are going in the direction we want. With JEV, we ran and compared our own implementations.
vooy team
We'd like to work with people who, when they discover new technology, build with it alongside their own agents and leave work that teammates can pick up.
vooy
Are you looking forward to new teammates too, Future Devil?
Future Devil
Bring your agents along. Let's build together and check the results. The future rules!!!
Design how people delegate work to AI and how people and agents share conversations. Take that work from problem definition through prototypes, user validation, and implementation.
Turn problems found through user research, quality assurance, and onboarding into product improvements. We value the drive to build AI skills and finish the work you take on more than your background or experience.
3–8 years of relevant experience. Lead financial operations and investor reporting, with specialist support for accounting and tax matters. Bring strong finance and accounting fundamentals and practical experience improving work with AI, then expand into business analysis and strategy.
For applications, questions, or referrals, email kimmy@vooy.com.
