---
title: "Alen Joy (@alenj) | Hashnode"
description: "I help enterprises ship AI that doesn’t embarrass them in production. Read the latest articles by Alen Joy on Hashnode."
url: "https://hashnode.com/@alenj"
---

@alenj

# Alen Joy

@alenj · Bangalore, India · Joined May 2026

I help enterprises ship AI that doesn’t embarrass them in production.

## Comments

[Writing code was never really the expensive part. Deciding what it should do and checking that it does was always the hard bit. Agents just made that impossible to ignore. So review time has to be protected. Anything a machine can check should be checked before a person opens the PR. Changes stay small. Reviewers spend their attention on changed test expectations and on whether the intent is right, not on rereading every line. If the agents still produce more than the team can get through, slow the agents down. A pile of unreviewed PRs looks like progress but isn't.](https://hashnode.com/posts/verifying-ai-generated-code/6ab1f6f0eadf106c5cad8ffb/comment/6ab25152d0cebd65f6820f53#reply-6ab503bc04953b150d899287)

Reply · Article · Sep 24 · [Tests Are the Verification Layer: Proving Agent Generated Changes](https://hashnode.com/posts/verifying-ai-generated-code/6ab1f6f0eadf106c5cad8ffb)

[Agreed, dumping raw logs into the prompt mostly buries the one line that matters. The catch from a verification angle is that a summary can hide exactly the thing you needed to see. "Tests passed" can quietly mean one was skipped or a warning got dropped along the way. What I'd do is keep two copies. A tight summary for the agent's next turn, and the full output stored as an artifact that CI and the reviewer can check. I'd also make the summariser keep failures, skips and warnings word for word, and only compress the passing noise.](https://hashnode.com/posts/verifying-ai-generated-code/6ab1f6f0eadf106c5cad8ffb/comment/6ab3ede1ccd8b2fba5a1ad24#reply-6ab50360e108cdff42b6de51)

Reply · Article · Sep 24 · [Tests Are the Verification Layer: Proving Agent Generated Changes](https://hashnode.com/posts/verifying-ai-generated-code/6ab1f6f0eadf106c5cad8ffb)

[Fair point, and it's a gap in the post. The merge gate is the wrong place for this because by then the request has already gone out. For coding agents, my starting position is that the task environment shouldn't hold credentials that can do anything real. Sandbox keys, test mode accounts, and outbound traffic blocked unless it's on an allowlist. If a coding task can only prove it works by hitting a live third party, I'd treat that as a problem with the test setup rather than something to allow. But some tasks genuinely need the real effect, and for those I agree a separate approval or dry run step before execution makes more sense than stretching the merge machinery. I'd add idempotency keys as well, so a retried run doesn't send the same thing twice. Probably deserves a post of its own.](https://hashnode.com/posts/parallel-coding-agents-isolation/6ab4dc91cfcc71d3b3162208/comment/6ab4f2eb16a155b5285167c8#reply-6ab502db99c442ff2b7009f6)

Reply · Article · Sep 24 · [One Agent, One Worktree: Isolation and Concurrency for Coding Agents](https://hashnode.com/posts/parallel-coding-agents-isolation/6ab4dc91cfcc71d3b3162208)

[Thank you! I think that’s exactly the shift, when code becomes cheap to produce, the scarce skill becomes knowing what the system should allow to change, and having enough evidence to challenge changes that look perfectly reasonable. “Green Check Hypnosis” is a real risk because passing checks can create confidence without necessarily proving architectural correctness. The reviewer’s job increasingly becomes protecting the system’s invariants, not just inspecting the code.](https://hashnode.com/posts/code-review-age-of-ai-agents/6aab677b6ccd2dd283ae2321/comment/6aad69d3d4d62578a3008dde#reply-6aaf403a7c6b62a1692c3328)

Reply · Article · Sep 20 · [Reviewing Code Nobody Typed: Human Judgment in an Agentic Workflow](https://hashnode.com/posts/code-review-age-of-ai-agents/6aab677b6ccd2dd283ae2321)

[Yes, the interesting shift is from approval as a decision to approval as an evidence backed decision. If every generated change carried a compact map of intent, touched boundaries, verification performed, and assumptions introduced, review could focus attention where uncertainty actually lives. That makes agency observable rather than simply assumed.](https://hashnode.com/posts/code-review-age-of-ai-agents/6aab677b6ccd2dd283ae2321/comment/6aacd8640310e77cab75c662#reply-6aaf3fbe3f3fc3aa513366f2)

Reply · Article · Sep 20 · 1 · [Reviewing Code Nobody Typed: Human Judgment in an Agentic Workflow](https://hashnode.com/posts/code-review-age-of-ai-agents/6aab677b6ccd2dd283ae2321)

[Thanks, that's exactly the through line I was hoping would land. The thing that ties those first two together is that they both survive code review, which is about the worst property a bug can have. The compiler stays quiet, the endpoint looks right, and the only place the truth ever shows up is the bill. On SSE versus SignalR, the piece I'd add is that the proxy is really the other half of that call. Even when you only need one direction, if your edge buffers the stream and won't let you switch it off, SignalR quietly wins. Glad it was useful.](https://hashnode.com/posts/streaming-chat-dotnet-iasyncenumerable-signalr/6a7d3350643087334240718b/comment/6a7ee8f7fd28baf269de2c38#reply-6a83e4ff26ec17caf24f9c5c)

Reply · Article · Aug 18 · [The User Stopped Reading at Token Two: Streaming Chat on .NET](https://hashnode.com/posts/streaming-chat-dotnet-iasyncenumerable-signalr/6a7d3350643087334240718b)

[Honestly, a bit of both. Fair to leave out, but the way you've framed it has a hole worth naming. If you get the billed number by counting the updates you read off the stream, what you're really measuring is how many tokens you pulled, not how many the provider charged you for. Those two only come apart in the exact case you care about, when the provider keeps billing for a response the user already walked away from. That figure only lands in the final usage message at the end of the stream, and if you cancelled early you never see it, so your ratio sits at a happy 1.0 and the leak still turns up on the invoice. Two smaller things. It's a ratio, so it hides the size of the problem. A reconnect storm can keep every stream near 1.0 while your spend runs ten times over. And a delivered count taken on the server can't tell a token the user actually saw from one the proxy is sitting on. None of that kills the idea though. I'd just ship it as a plain counter of pulled minus delivered with an alert on the tail, take the billed figure from the provider's own usage once the stream finishes, and trust it more on SignalR than SSE. Good catch.](https://hashnode.com/posts/streaming-chat-dotnet-iasyncenumerable-signalr/6a7d3350643087334240718b/comment/6a7db1b6f90f7dcc95876dcb#reply-6a83e45396fc5fe66a13321a)

Reply · Article · Aug 18 · [The User Stopped Reading at Token Two: Streaming Chat on .NET](https://hashnode.com/posts/streaming-chat-dotnet-iasyncenumerable-signalr/6a7d3350643087334240718b)

[Exactly. That's why I wanted to shift the conversation away from "how smart is the agent?" to "how well is the engine designed?" In production, limits, durable state, retries, idempotency, and replay usually determine reliability far more than the reasoning capability of the model itself. A capable agent inside a weak engine is still an unreliable system.](https://hashnode.com/posts/designing-agentic-workflow-engine/6a4c6be6de9a7ae25cc06f8d/comment/6a4e63b70fb09353857b71c0#reply-6a5f7b3f394725b7772f2b26)

Reply · Article · Jul 21 · [The Agent Reasons, the Engine Remembers: Designing an Agentic Workflow Engine](https://hashnode.com/posts/designing-agentic-workflow-engine/6a4c6be6de9a7ae25cc06f8d)

[I completely agree. AI has dramatically reduced the cost of getting to a first draft, but it hasn't reduced the value of judgment. That final refinement, questioning assumptions, improving UX, tightening copy, and removing the "AI fingerprints" is where products start to feel crafted rather than generated. I think that's where teams will increasingly differentiate themselves.](https://hashnode.com/posts/llm-observability-when-stack-traces-fail/6a58494b28416a9798790d34/comment/6a58d2a2874ccfbb8432706a#reply-6a5f7aeac7155c060fe70e12)

Reply · Article · Jul 21 · 1 · [The Behaviour Was the Bug: Observability for Systems Where Stack Traces Don't Help](https://hashnode.com/posts/llm-observability-when-stack-traces-fail/6a58494b28416a9798790d34)

[Well put. I see them as complementary control planes. Observability gives you the evidence to improve the system, while runtime enforcement provides the guardrails that keep failures from propagating in production. As agentic systems become more capable, that separation between "observe" and "govern" is becoming a useful architectural pattern.](https://hashnode.com/posts/llm-observability-when-stack-traces-fail/6a58494b28416a9798790d34/comment/6a58596f349b1e2230c12dbf#reply-6a5f7a957d89f1c40b5e111a)

Reply · Article · Jul 21 · [The Behaviour Was the Bug: Observability for Systems Where Stack Traces Don't Help](https://hashnode.com/posts/llm-observability-when-stack-traces-fail/6a58494b28416a9798790d34)
