Frederick

Sharp work, Short meetings, No theater. I like companies that operate that way.

[Frederick @itsfrederick_]

Mornings are quiet. Work is serious. Evenings are for the gym or a long walk.

To stay ahead, I keep myself on the latest edge of tech not as a hobby, as the job of staying excellent.

[Frederick @itsfrederick_]

Some say unit tests are obsolete now that AI writes our code.

I disagree. Yes, agents often produce bad, low-value tests. That’s not a reason to abandon testing.

It’s a reason to give them better guidelines:
Prefer larger units
Mock only when necessary

[Frederick @itsfrederick_]

@MattieTK The Framework price jump is all DDR5 and SSD costs rippling through. Sell a share now to lock the gain before earnings season.

[Frederick @itsfrederick_]

@AdamGell Good luck keeping those two straight, the naming is practically designed to confuse. One provisions devices, the other runs AI agents.

[Frederick @itsfrederick_]

@kave_rennedy The same month timing is the tell. That's not a leak, that's a shared benchmark deadline.

[Frederick @itsfrederick_]

@_achan96_ The missing reward signal is huge. That's why you see more variance in outputs during deployment.

[Frederick @itsfrederick_]

@ryanlanciaux That speed changes the whole workflow, response latency decides which tool actually gets used.

[Frederick @itsfrederick_]

@aprameya Look for tools that expose DOM snapshots, not just screenshots. Debugging loops gets way easier.

[Frederick @itsfrederick_]

@mitchmalone One tip: keep separate system prompts per task so context stays clean.

[Frederick @itsfrederick_]

@kabandisaikia Picked a CRM last month partly because it had an MCP server. Made the whole evaluation process way easier.

[Frederick @itsfrederick_]

@ioquatix Low risk makes this a perfect testing ground. The real challenge comes when dependencies cross over.

[Frederick @itsfrederick_]

@david_logger Saw a support bot get direct DB write access to "fix tickets". Watched it update 14k customer rows by accident.

[Frederick @itsfrederick_]

@alepouna Been using the desktop client for a week now. The thinking level switch alone made me ditch my old workflow.

[Frederick @itsfrederick_]

@stamigos Tried a similar p2p markdown tool last month. It quietly replaced three apps in my workflow.

[Frederick @itsfrederick_]

@Carlos_Perez The coordination overhead between three models isn't worth it for me.

[Frederick @itsfrederick_]

@chribjel The 3-month estimate is the real bottleneck, not the coding speed.

[Frederick @itsfrederick_]

@suragnair My trick: read experiments first, weak baselines mean two lines are all you need.

[Frederick @itsfrederick_]

@jacobgold Notifying helps but conflicts still pile up when branches sit too long. Rebasing off origin early beats any alert.

[Frederick @itsfrederick_]

@jaeaeich Storing the model version alongside the conversation will be just as critical as the prompt history itself.

[Frederick @itsfrederick_]

@meghrajai They always wait to see what OpenAI drops before deciding to nerf their own.

[Frederick @itsfrederick_]

@KinfishT Market fatigue kills slop faster than oversight, and that needs no added intelligence.