The operating problem
Customer context sits in a CRM. Team updates sit in chats. Goals sit in documents. Deadlines sit in someone's memory. Leadership becomes the integration layer, chasing information and manually deciding what should happen next.
The system
I designed bwork.space as an AI-native operating workspace. It connects contacts, customer relationships, projects, tasks, goals, risks, team accountability and communications so the system can form a usable picture of the work.
- Workspace-specific projects, goals, tasks and ownership
- Operational updates drawn from live work instead of duplicate reporting
- Risk and blocker detection
- Scheduled management and accountability loops
- An Autonomy layer that evaluates possible actions against policy
- Communication workflows connected to actual work records
Why this is not an autoresponder
An executive system should not merely turn every event into a reply. It should determine whether the event needs action, whether evidence is sufficient, who owns the next step and whether execution is safe. Sometimes the correct result is a message. Sometimes it is a task, a risk, a meeting proposal, a system change or no action at all.
Guardrails
Low-risk and reversible actions can move automatically. Financial, legal, destructive, security-sensitive and commitment-making actions remain review-first. The objective is useful autonomy with accountability, not automation theatre.
My role
I lead the product direction, architecture and engineering. I also operate the system against real Raysource work, which means failures appear quickly and cannot hide behind a demo.
Current engineering lesson
The hardest part of agentic software is not producing language. It is grounding decisions in current state, choosing the correct action type, executing through narrow interfaces and reporting the actual outcome. That is where most of the engineering work now sits.
Visit bwork.space
Founder and CTO of Raysource Labs. I build AI products, workflow automation and operational software.