Where we’re going
New ways of working have always given rise to new tools. Agile reshaped how teams plan and ship, and a toolchain like Jira made it routine. Human–agent collaboration stands at the same point — and the tools that carry it don’t exist yet. That gap is what Looply builds.
The market moment
We start with SaaS and e-commerce engineering teams larger than five people — where several agents already ship production code against live customers, coordination breaks first, and a defect is most expensive. From there the same workflow extends to every kind of work agents do by writing code.
Because we price per active engineer, the opportunity grows as AI decouples output from headcount — and the bands below are sized bottom-up, at $360 per engineer per year.
The trajectory
Two phases on one workflow — the same shared standards, memory, and evidence, applied to a widening surface of work.
A team connects a repository and agrees its standard once. People and agents ship code inside one workflow — intent and decisions carried across every handoff, each result returned with evidence, and the accountable leader accepting or sending it back.
Agents increasingly carry out work well past software — shipping orders, investigating a revenue discrepancy — by writing code. The same workflow governs it: a person sets intent and standards, agents coordinate, and the leader judges the business result.
We deliberately don’t build toward unattended, no-human hand-off. Keeping a person accountable for the result is a permanent principle, not a way-station — AI doing far more of the work, a person still standing behind what ships.
Why it holds
A tool is copyable; an accumulated way of working is not. Every result a team accepts or corrects becomes memory — decisions, exceptions, and evidence tied to the standards that produced them — and Looply learns from it. That is the moat, and it grows with use.
The standard and the memory belong to you, not to us.
What we believe
AI produces work faster than people can check it, and teams now run several agents at once. Assumptions pass silently between them, parallel tasks collide, and decisions get lost between sessions. The old workflow was built for human execution and human review — that world is gone.
A model grading its own output is the fox guarding the henhouse. However good a model gets, someone still has to see how a result affects function, security, and structure before it’s accepted. The author can never be the one to certify that.
What “good” means belongs to the team, not to each tool’s config or each engineer’s head. Written down once, owned by the team, and applied at the moment of acceptance, a standard holds across every person and agent — and survives a change of model.
Not as a stopgap. As a principle. On payments, on core systems, everywhere. People set intent and standards, agents coordinate and do the work, and the accountable leader accepts the outcome — with the evidence to stand behind it.
The end state isn’t more output. It’s a team whose standards, memory, and methods get sharper with every result — so instructions need less elaboration, the same mistakes stop recurring, and each task starts closer to what the team meant.
Build it with us
Free for individuals at launch, per-seat for teams from March 2027. We want the people who share the conviction early.