Kevin

🧑🏼‍💻Software Developer passionate about Javascript & people || 🔻 ex-host of the @roromelb || Dad

[Kevin @optimistic_updt]

🔥 Hot Take:

Study
- **human communication** &
- **behavioral science**

instead of prompt engineering 🧐

You ll be a much better AI Engineer 😲

[Kevin @optimistic_updt]

Harnesses are the new IDEs & more. 🚀
I think you'll see as many agents harness as you have people in your company.

Keep in mind:
- Code is the end result, agree on that
- Provide a certain floor so that you can speed up the laggard

[Kevin @optimistic_updt]

There's something so satisfying about cleaning up your logs and color coding them.

The terminal becomes like the most beautiful movie...

[Kevin @optimistic_updt]

https://t.co/uK1il7k6LL

[Kevin @optimistic_updt]

💡
Been achieving much better results by aligning with agents upfront and thoroughly exploring every aspect of a request or decision before spending too many tokens on creating things

I often use a variety of techniques, but "grill-me" from @mattpocockuk skill is a great start

[Kevin @optimistic_updt]

🔍POV:
You thought your agent's biggest hurdle was big data 📈.

Turns out it's deciphering the cave-painting the user just smashed into the chat...😰

[Kevin @optimistic_updt]

When did buying flights became such a head fuck 🤯

You have less chance of getting scam buying Bitcoin these days

WTF

[Kevin @optimistic_updt]

Just delivered a massive feature. 🔥

My self-assigned reward?

Spend a day cleaning up logs 🫦

[Kevin @optimistic_updt]

Going into the day like so...

[Kevin @optimistic_updt]

I guess the Australians will be running things at night....

[Kevin @optimistic_updt]

We don't get to ship and forget.

The responsibility to protect end users sits with the people building the tools.

[Kevin @optimistic_updt]

LLMs are trained on the dumpster fire that is the internet. @Rathness37 put it plainly at @DDDMelb

Bias isn't a future problem. It's baked in right now, it's shipping to your users every time you deploy without guardrails

As builders, that's on us!

Three things you can do 👇

[Kevin @optimistic_updt]

3. AI reviewing AI — Use a second model as a quality gate. It's not foolproof, but it catches what you'll miss.

People have died from this one!

[Kevin @optimistic_updt]

Check him out at https://t.co/jBsgMCz3Iy

[Kevin @optimistic_updt]

Autonomy earns trust, makes better decisions but is slower.
You get to spend the trust earned to make fast decision with authority, but be careful: there is more chance of being wrong.

Fantastic talk from Andrew Murphy at @DDDMelb

[Kevin @optimistic_updt]

@claudeai finally 🥵

[Kevin @optimistic_updt]

https://t.co/jp9lBgZ4ni

[Kevin @optimistic_updt]

"In a system where agent throughput far exceeds human attention, corrections are cheap, and waiting is expensive."

Ryan Lopopolo, OpenAI