Active-task carousel
Bring the next task that needs human action forward while other work continues.
IMPROVE THE WHOLE DEVELOPMENT SYSTEM
Make Human + AI
more productive
than AI alone.
You cannot improve a process you cannot measure. Midva records each step and handoff, along with the time and token cost. The record shows the human where to reduce waiting and move more useful work through the system.
OBSERVED TASK / INSIGHT 001
Six hours of work. Sixteen hours to finish.
HUMAN
01:00
1
AI
02:00
2
HUMAN
01:00
3
WAIT / HUMAN
09:00
TASK START
DEPLOYMENT
- Lead time
- 16:00
- Active work
- 06:00
- Flow efficiency
- 37.5%
- Handoffs
- 3
Constraint / Human. Three AI handoffs produced three hours of AI work. The task waited ten hours for human action, including a nine-hour overnight wait before the final review and deployment. Under Goldratt’s Theory of Constraints, that human wait limits the throughput of the whole system.
Lead time is elapsed time from the task start to deployment. Throughput is useful completed work per unit of time.
Read Insight 001Midva comes from the dual, a Slovene grammatical form for exactly two. It was voted the most beautiful Slovene word and means "we two," or you + I. Here, the pair is you + [A]I.
The word treats both members as one pair. Midva applies that idea to the development system.
Measure the constraint. Reduce the wait.
In a Human + AI pair-programming system, the human is the constraint: the pair’s throughput depends on how effectively limited human attention is used. These 4 feature examples serve the same system goal: reduce switch cost and human waiting while making each human action more effective.
Handoff signals
Sound and visual notifications make a finished AI run hard to miss.
Async coordination
Messages, comments, and queues let human work advance without synchronous actions after waiting.
Diff reviewer
Write inline or batch comments directly to the AI actor, then hand the reviewed work back without another coordination loop.
Results from one working system.
Midva is currently used by one developer working primarily with Elixir, Phoenix LiveView, and JavaScript hooks. The starting point was Human + AI pair programming; today that pair runs through the current measured process.
- Estimated throughput change
- 2.5×
- Observed working pace
- 8.8tickets / 10 working hours
- Human time utilization
- 92%
In Midva Blog, the owner shares what he learned while building a measured Human + AI development system.
INSIGHT003
To switch or not to switch?
A synchronized timeline shows why keeping the human constraint busy can improve throughput while delaying the first individual result.
Open Midva Blog
Midva is owner-only for now. Interested? Write to Oleksandr on LinkedIn.