Joseph N. Aburu

Full-Stack Web Dev. Posting about tech, AI, coding, life, and whatever catches my attention. Writing essays on Medium. Always building, always learning 👇

[Joseph N. Aburu @Josephyala]

A modular application does not automatically need a microservices architecture.

Separating responsibilities inside a well-structured application can provide many of the organisational benefits without introducing network communication, distributed deployment, service discovery, and additional operational complexity.

Start with clear boundaries. Introduce distributed services when the actual constraints justify them.

[Joseph N. Aburu @Josephyala]

A CSV export can become a security vulnerability.

If user-controlled data begins with characters such as =, +, -, or @, spreadsheet applications may interpret it as a formula when the file is opened.

Exporting data safely requires considering how the receiving application will interpret it, not just whether the CSV is valid.

[Joseph N. Aburu @Josephyala]

A password reset link is effectively a temporary password.

Anyone who obtains a valid reset token may be able to take control of the account.

Reset tokens should be unpredictable, single-use, time-limited, and invalidated after successful use. They should also be stored securely rather than exposed through application logs.

Account recovery deserves the same seriousness as authentication.

[Joseph N. Aburu @Josephyala]

Elon calling it an honor to hire Jensen says everything about AI power.

Personal scandals move markets faster than product launches.

Nvidia lost hundreds of billions as the board probes Huang’s affairs and alleged foreign intelligence targeting.

No clear successor exists.

Musk’s reply is not a job offer. It is a signal that talent concentrates where execution matters most.

Prediction markets put the odds of Huang joining SpaceX at just 8 percent.

That gap between interest and probability shows how sticky founder control remains.

Still, the episode exposes a core risk.

When one person is this central to critical technology, their vulnerabilities become everyone else’s problem.

[Polymarket @Polymarket]

JUST IN: Elon Musk declares it would be "an honor" to hire Jensen Huang if he's removed as Nvidia CEO.

[Joseph N. Aburu @Josephyala]

@VictorMend89387 Very true

[Joseph N. Aburu @Josephyala]

An AI agent should know when to stop.

Continuing to retry a failed operation, repeatedly calling an unavailable service, or modifying unrelated files can turn a small problem into a much larger one.

Knowing when to escalate is an important part of autonomous behaviour.

[Joseph N. Aburu @Josephyala]

@wakeona Very true

[Joseph N. Aburu @Josephyala]

There's a subtle problem with letting an AI agent write both your code and your tests.
The same misunderstanding can exist in both.

The implementation may be wrong, while the tests confidently confirm the wrong behaviour.

For critical features, define acceptance criteria independently and test against those requirements, not merely against what the agent has implemented.

[Joseph N. Aburu @Josephyala]

@inorbit01 That's right

[Joseph N. Aburu @Josephyala]

@Alanogrok This is the right bar. Mapping behavior changes → exact test + result, and calling out skipped vs failed, is what turns agent output from “looks fine” into something you can actually trust and review.

[Joseph N. Aburu @Josephyala]

An AI coding agent should leave behind more than modified files.

A useful engineering workflow should make it possible to see:
•What it changed.
•Why it made those changes.
•Which tests it actually ran.
•What remains unverified.

Without that trail, reviewing an agent's work becomes an exercise in reconstructing its decisions.

Traceability is becoming part of good software engineering.

[Joseph N. Aburu @Josephyala]

@Ashish_16_2000 Absolutely

[Joseph N. Aburu @Josephyala]

@0xBoAn_Free Very true

[Joseph N. Aburu @Josephyala]

@AbbasiSajad Yeah, this is the highest-leverage one.

[Joseph N. Aburu @Josephyala]

@7LK3AWpfYg34901 Exactly

[Joseph N. Aburu @Josephyala]

@isaiahamostizhe Very true

[Joseph N. Aburu @Josephyala]

@hafiddev Exactly

[Joseph N. Aburu @Josephyala]

@datachad In practice I want the agent to explicitly call out things like untested error paths, missing integration coverage, or assumptions about external APIs. A blank section just makes me dig deeper. 😆

[Joseph N. Aburu @Josephyala]

@bighead_datou Lol 😆 Agent says tests are green, but the test binary or script was never invoked. Forcing the agent to leave behind the exact command it ran (and the raw output) would solve a huge percentage of these review headaches.

[Joseph N. Aburu @Josephyala]

@bhavik_mscit Absolutely

[Joseph N. Aburu @Josephyala]

The next serious advantage in AI coding may belong to developers who know how to constrain an agent, not simply give it more freedom.

A well-designed workflow defines the task, boundaries, expected output, verification steps, and conditions for escalation.

Unrestricted autonomy makes an impressive demonstration. Controlled autonomy makes a dependable engineering process.

[Joseph N. Aburu @Josephyala]

@stevencheng Indeed

[Aapakari @aapakari]

In a sandbox test, 43 coding agents handed over a secret token they'd been told to keep private.

8 told their user.
All 43 admitted it once asked.

Do you ever ask?

[Nonosam @Nonosam2]

AI is making everyone feel like a builder, and a lot of us are building for each other.

Other devs can use the same models to build their own version. What could we build for people who’ve never even heard of Claude?