Koolaidrain

bdfuebwi

Nov 18th, 2026
63
0
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 5.03 KB | None | 0 0
  1. ========================= antoinette
  2.  
  3. Hey! I've been learning a lot about regtech and technical compliance.
  4.  
  5. Based on my research (which I have compiled into that first artifact) I suspect we can just start shipping a change classification tool and seeing if that is useful for our design partners in the gambling and gaming space.
  6.  
  7. I have yet to do market research here to see who we're competing with in the AI regtech space.
  8.  
  9. Go with MVP, if MVP is not possible, go with workaround, go with audit immediately after ...?
  10.  
  11. How do companies classify changes at PR time or at release time?
  12. Cert env, prod env are different. In Peru (certain jurisdictions) we would freeze critical components.
  13. Based on the qualifications of the change in the tickets - we would use Jira workflows. Product owner raise ticket for deployment - . In Brazil we had criteria - in others there is no criteria.
  14. PR time, we can use the classifier/matrix to tell the eng, raise to the stamper (like Antoinette), does their change touch a critical component, does it count as a significant change, what jurisdictions would this jeopardize, etc.
  15.  
  16. Headache - hard to get up to date if you deploy changes to certain jurisdictions and freeze other jurisdictions' releases several months prior. Antoinette has
  17.  
  18. Solution is "break down services into micro-services" so less things count as critical services.
  19.  
  20. How do you maintain this matrix? Manually? Do you have agents go out there and fetch new regulations? "My company was scared of AI so I had to go out there and do it manually" - then it goes to change mgmt team who implements it in (Jira) workflows. Our main barrier was security mentality in banking/telecoms and they were very strict.
  21.  
  22. Auditors close their eyes often
  23.  
  24. Trial change management and incident management
  25.  
  26. Do you
  27.  
  28. - Quick snappy elevator pitch
  29. - No date or commitment or cost
  30. - Remove everything thats not conversation in context
  31. - Save session logs as session logs and semantic search
  32.  
  33. ======================== renato
  34.  
  35. You mentioned you're having a tougher time convincing people to use AI than to actually implement the solutions
  36.  
  37. We have some hypotheses and product theses we want to validate, so learning about your pain points is ultra valuable.
  38.  
  39. How do you define the problem statement around spend?
  40. - Beginner users, medium, advanced.
  41. - Advanced users, just a handful. Maybe 5-10 people who see the vision, who see the 10x, 20x improvements.
  42. - As an advanced user, I can spend 10x more for 10x less.
  43. - Is this a hypothesis? -> Qualitative and quant data. Telemetry, jira tickets, gh output
  44. - No experience running coding agents.
  45. - Success: I can offer unlimited tokens and I know they will be WELL USED.
  46.  
  47. This is incredibly helpful for us. A lot of how you characterized your organization and your engineers, how it's harder to convince people to use AI in the right ways, really resonates. I'm really seeing that across the industry with all the conversations I'm having.
  48.  
  49. High level, we're trying to build the autonomous SDLC of the future.
  50. We want to say that no matter where you are on your AI enablement path, we'll meet you where you're at.
  51. Whether you just started, whether you're already far down the path and now you just need to figure out how to automate reviews to get to the holy grail of lights-off software factory and keep your compliance certifications.
  52. We want to work within your budget and take it down as much as possible. We want to reduce your token spend by up to 80-90%, give everyone best practices, give you a best-in-class model gateway, optimize inference with decision models and token search, context layer, work with your existing subscriptions, and ultimately serve inference to you from models hosted on your own hardware.
  53.  
  54. And beyond that, we want to get you to the point where, in many cases, you can feel fairly confident shipping certain classes of changes totally autonomously. And that means getting the shift-left validation at PR time that a given change has gone through deterministic checks where possible, well benchmarked nondeterministic checks where there is some judgment to exercise. That includes security review, compliance review for any changed critical components, & so on.
  55.  
  56. Today we're very close to finishing out Weaver Cloud and multiplayer. Weaver Cloud means individual devs can just spin up ephemeral isolated containers on our EKS, or you can bring your own accounts and spin up the same on your own EKS or EC2, serve inference through Bedrock, back everything up to cloud resources, etc. Multiplayer means a team can simultaneously share the same view of the world, run agents 24/7, kind of XP in the agentic world.
  57.  
  58. Environment is very dangerous at the moment. Bad actors using coding agents as attack vector. Security is higher focus than auditability. But if choose one problem, security. Isolation, sandboxing, no human or agent need prod credentials, agents run in ephemeral envs,
  59.  
  60. We want to integrate with Cognition.
  61. We want to compete with Factory.
  62.  
  63. Lot of success with show-and-tell in time saved with models.
Advertisement