Ricardo Torres

Strong opinions, loosely held because I'm probably wrong.

[Ricardo Torres @riffingonsw]

I find it exciting.

[Ricardo Torres @riffingonsw]

Software is changing. You won’t be left behind unless you want to be. Wholly new ideas might take over. Your old knowledge is still valuable. It’s hard because all of this, and more besides, can all be true at the same time.

[Ricardo Torres @riffingonsw]

@thorstenball Not just rethink, reimagine and reinvent.

[Ricardo Torres @riffingonsw]

We’re still not thinking big enough.

[Thorsten Ball @thorstenball]

More thoughts on jellyware & why engineers turning their nose up at it might be wrong in the long run

[Mitchell Hashimoto @mitchellh]

Been fighting the Great Firewall of America all day (Fable too, though this screenshot is from Codex). I truly believe this is resulting in software being less secure because we can't find and fix security issues in our own software that we control without a bunch of red tape.

[Ricardo Torres @riffingonsw]

As an MCP detractor, it’s not that MCP is bad. (It could be better, but what couldn’t?) It’s that there are a lot of bad MCP servers. On the positive side, this does seem to be getting better.

[Ricardo Torres @riffingonsw]

The models themselves are an evaporating moat.

[Ricardo Torres @riffingonsw]

@hvent90 But again, early days.

[Ricardo Torres @riffingonsw]

@hvent90 Then there’s the satellite/mothership idea. My local agent is separate from and shares with my long-lived agent.

[Ricardo Torres @riffingonsw]

@hvent90 Memory allows progress. Progress leads to skill. Skills should be codified, where possible.

[Ricardo Torres @riffingonsw]

@hvent90 Early days, but basic idea is about long lived agents doing self-improvement (Hermes) with durable memory (Letta Code) but also being able to modify their own code in a supported, extensible manner (Pi).

[Ricardo Torres @riffingonsw]

@hvent90 So, Pi, Hermes, and MemFS meet in a bar…

[Ricardo Torres @riffingonsw]

@hvent90 It’s funny because my latest harness is trying to go the other way. After starting and stopping more than 5, I think the problem is I want an agent, not a harness, but we’ll see. Lol

[Ricardo Torres @riffingonsw]

Started work on my sixth or seventh harness this week. (Most got to a barely working state before being put aside.) This morning I find myself wondering if the issue is I don’t want a harness. I just want an agent.

[Ricardo Torres @riffingonsw]

@glcst @theo Took one last time I flew into SJC. At that moment, it was cheaper than an Uber, but I did have a human driver (probably Uber, lol) turn right from the left lane in front of us, cutting us off. Pretty clear the human knew the Waymo would yield, lol.

[Ricardo Torres @riffingonsw]

@hvent90 I haven’t had a chance to sit down with the new MCP RC. I’m hearing good things, but to your point, I think a lot of MCP servers were bad due to implementation, not the standard.

[Tomás Senart @tsenart]

Starting to map the graph for a founding engineer at Perfloop.

Looking for real systems depth - perf, infra, observability, AI infra, devtools, databases, runtimes - plus extreme agency, product/architecture judgment, and native agentic building.

Europe / near-Europe overlap preferred.

[Ricardo Torres @riffingonsw]

@glcst Lol, such a troll.

[Ricardo Torres @riffingonsw]

Context is key. It’s key for humans. It’s key for AI agents.