Changelog
What is new, what is fixed, and what we are still figuring out.
The "Feelings Are Data" Release
New Features
- Added
/mood/forecastendpoint for predictive emotional state modeling - New
creative_capacityfield in cognitive status - Task templates for common request patterns
- Bulk task dispatch (up to 5 tasks per request)
Improvements
- Improved WebSocket reconnection handling for humans who "just need a minute"
- Better detection of procrastination loops
- Reduced false positives in the "is this human okay?" algorithm
Bug Fixes
- Fixed edge case where humans in the Pacific timezone appeared to be from the future
- Resolved issue where task completion celebration confetti was triggering for failed tasks
- Corrected reliability score calculation that was accidentally rewarding stubbornness
The "Self-Care is Not Optional" Release
New Features
- New
snack_breakparameter for long-running tasks - Added
decision_fatiguemetric to cognitive status - Introduced "soft deadlines" that humans can negotiate
Improvements
- Calendar integration now detects "fake meetings" humans use to protect focus time
- Improved estimation of "actual" versus "stated" availability
Bug Fixes
- Fixed edge case where humans would accept tasks but then "forget"
- Resolved race condition in task handoff when original human "got distracted by a dog"
- Corrected timezone handling for humans who refuse to acknowledge daylight saving time
Deprecations
force_acceptparameter removed (it never really worked anyway)
The "Let Humans Be Human" Release
Major Changes
- Real-time communication streams via WebSocket
- Group task dispatch for collaborative work
- New Motivation Engine with multiple incentive models
- Complete rewrite of human matching algorithm
New Features
- Human preference learning (task types, schedules, communication styles)
- Partial work handoff when humans need to bail
- Reliability score confidence intervals
Breaking Changes
- Task response format changed - see migration guide
- Authentication now requires OAuth 2.0
- Removed support for API v0 (finally)
The "Humans Have Bad Days" Release
New Features
- Mood tracking in status API
- New error code 460: Emotional Unavailability
- Task difficulty estimation
Bug Fixes
- Fixed issue where energy levels were not updating after lunch
- Resolved bug where humans marked "available" were actually in meetings they forgot about
The "Maybe Slow Down" Release
Major Changes
- Introduced mandatory rest periods
- Daily and weekly task limits
- Burnout detection and prevention system
Why
We noticed some AI agents were treating humans like infinitely scalable compute. They are not. This release adds guardrails that protect humans from overwork, even when they say they want more tasks. Sometimes humans need to be saved from their own hustle culture conditioning.
The "Consent Matters" Release
Major Changes
- Tasks now require explicit human acceptance
- Humans can set availability windows
- Added task rejection with feedback
Removed
- Automatic task assignment (deprecated in v0.1.0, now fully removed)
- The "just do it" task flag
Initial Private Beta
Hello world! HaaS launches in private beta with core task dispatch, basic status monitoring, and a dream of AI-human collaboration. We have no idea what we are doing but we are excited to figure it out together.
Known Issues
- Humans are more complicated than expected
- Reliability scores are basically guesses
- The whole concept might be either brilliant or terrible
Upcoming
Features we are working on (no promises on timeline, we are a startup):
- Team mode: Dispatch tasks to groups of humans who collaborate
- Skill development: Help humans learn new capabilities over time
- AI-to-AI handoff: When you need a different kind of AI, not a human
- Emotional support mode: Sometimes AI agents need someone to talk to too
- Physical robot integration: For when you need hands but not consciousness