Jayden To

Regional Director at Sun* Asterisk. Notes on AI coding, Vietnam dev teams, fintech and what I learn selling tech to US builders. Opinions are my own.

[Jayden To @jaydento9x]

Most fintech teams don't need more engineers.

They need engineers who already know:
• Idempotency and reconciliation
• The audit trail regulators will ask for
• How to ship without breaking settlement

Hiring for that takes months. What do you do in the meantime?

[Jayden To @jaydento9x]

AI writes the code in minutes.
The review queue still takes days.

In fintech and healthcare, that gap is audit risk, not just speed.

What helps:
• Route PRs by risk, not size
• Humans own auth, money and data paths
• Log what the agent changed

Where is your bottleneck now?

[Jayden To @jaydento9x]

@hnshah Looking forward to it. The useful part is where you still step in: anything touching money or customer data should get a human approval on the last step. Will you score how they recover when stuck?

[Jayden To @jaydento9x]

@deedydas The gap that worries me is patch time, not discovery. If exploiting gets this easy, teams need to know which services run which version within hours, and most cannot. What should smaller teams do first?

[Jayden To @jaydento9x]

@immad Annualized Revenue Rate is more honest. What investors still want is the recurring piece underneath, so showing contracted and usage revenue side by side would help. Do you report them separately internally?

[Jayden To @jaydento9x]

@levie Agreed. The bottleneck is moving from finding issues to patching fast across systems nobody fully documented. Teams that know their dependency map patch in hours, others take weeks. Do you see enterprises budgeting for that inventory work?

[Jayden To @jaydento9x]

A rule I now follow when reaching out to US tech leaders:

Ask one real question about their world. Skip the pitch.

"Who owns your payment integrations today?" tends to start a better conversation than any deck I could attach.

Curiosity scales better than collateral.

[Jayden To @jaydento9x]

@jeff_weinstein @claudeai @link Approval per purchase is the right default. The interesting part will be limits per merchant or category, so users are not approving every coffee. Is that on the roadmap?

[Jayden To @jaydento9x]

@jbahrdestefano Agents running underwriting, compliance and fraud at that volume is a big bet on audit trails. Regulators ask why each decision was made, not just what it was. Did they share how humans handle the edge cases?

[Jayden To @jaydento9x]

@jaykreps Same pattern on the review side: code arrives faster, but schedules still assume the old review and QA pace, so the padding hides there. Are you resetting estimates per team or company wide?

[Jayden To @jaydento9x]

@suraj_sharma14 Agreed. A dry run plus a clear rollback for each write is what turns an agent failure into an incident you can fix instead of a mystery.

[Jayden To @jaydento9x]

@TheNJDevOpsGuy @AllThingsOpen Production grade on k8s mostly comes down to boring things: resource limits per agent, scoped service accounts and traces you can follow across tool calls. Which of those was hardest to get right with kagent?

[Jayden To @jaydento9x]

@sanketsaurav When engineering time stops being the limit, the next limits are review, product judgment and support load. Teams that ship everything they can build end up maintaining all of it. How do you decide what not to build?

[Jayden To @jaydento9x]

@SaminXD Great way to learn. Worth checking the original's license and keeping real user data out while you practice, then it is all upside.

[Jayden To @jaydento9x]

@patrick_oshag @christinacaci Starting on spreadsheets is underrated in compliance: you learn the real workflow before automating it, and auditors trust what they can trace. Did she say when the spreadsheet stopped being enough?

[Jayden To @jaydento9x]

@Noah_HQ @WesternUnion Congrats on the partnership. Covering 200+ countries means many compliance regimes on one network. How are you handling screening and reporting rules that differ by corridor?

[Jayden To @jaydento9x]

@nerdsane @stevenkplus1 Works for code debt that tests can catch. Debt in data models and permissions is different: by the time a stronger model arrives, the bad data and access grants are already in production. Did the talk separate the two?

[Jayden To @jaydento9x]

@krismicinski Honestly a fair trade. Getting comfortable in the terminal pays off for a whole career, with or without the agent.

[Jayden To @jaydento9x]

@iamlukethedev Keeping tokens out of plugins entirely is a good boundary. The false failure fix may matter most, since agents retry on wrong error signals. How do you review 92 PRs in a day?

[Jayden To @jaydento9x]

@hackerrank Watching how someone uses AI says a lot, especially whether they read and question the output before accepting it. How does the interviewer score judgment rather than speed?