YATFA vs. Factory

Multi-agent by default, or multi-agent on demand

Both companies recognized that one agent doing everything fails; they put the cure in different places. Outside a Mission, a Factory session is one Droid doing the work and judging it. In YATFA the worker never reviews its own work — the reviewer is a distinct role with a distinct context, on every ticket, with nothing to launch first.

Determinism scoped to a goal, or as the always-on layer

This is not “YATFA has determinism and Factory does not.” Factory’s Missions runner is genuinely deterministic, and their own description of it is worth reading in full:

“A programmatic runner takes the feature list and spawns a worker for each feature in order. Each worker starts with a fresh context, receives its feature spec, writes tests first, then implements.”

Fresh contexts per feature, tests written first, and validators that judge without fixing — that is a well-designed system, and it is deterministic in exactly the way we claim to be. The difference is scope. Factory’s runs inside a Mission you launch; YATFA’s runs continuously across everything, including detecting stalled work, re-queueing it, and escalating to a human after a bounded number of failed review cycles. For a codebase you leave agents running on, that continuity is the meaningful reliability difference.

Improving itself, or improving your codebase

This is the sharpest difference, and it survives scrutiny. Signals is recursive self-improvement: the agent analyzing its own behavior and filing work to fix itself. It does not sweep your repository for tech debt. YATFA’s Researcher does exactly that — scanning for missing tests, stale documentation, and architectural problems, then filing structured proposals you approve or decline. Factory automates the agent’s backlog; YATFA proposes work on yours.

Memory: a real split, not a clean win

Against Factory’s local terminal agent the gap is large, and their own documentation says so plainly:

“Droid doesn't have built-in memory between sessions, but you can create a powerful memory system using markdown files, AGENTS.md references, and hooks.”

Note the second half: Factory’s answer is that you build the memory system yourself, and their docs show you how. Against Factory’s cloud platform the gap narrows further and the shapes differ — Factory offers hierarchical user and organization memory with a sophisticated retrieval stack, while YATFA’s is cross-agent by construction and self-updating in the loop, with nothing to assemble. Different shapes of remembering, not one present and one absent.