Light

Building fun things with AI. Exploring agents and sharing what I learn along the way.

[Light @opschainx]

The AI products that really work will need great user interaction.

Then come built-in tools, task scheduling, and delegation.

Every personal agent will need its own cloud computer and environment.

Voice models still have a lot of room to grow. Grok’s voice is damn good: smooth conversations, interruptions, and picking up where you left off.

Then there’s the agent desktop: permissions and seamless control of apps. There’s a huge market here, with so much to build.

[Light @opschainx]

Build time — agreed, and your reason is the right one. By the time a delete lands, a reconstructed stamp can't tell a changed record from an untouched one.

On the closure problem: Kairos dodges part of it by construction. A memory only gets in with a verbatim quote from the user's own turn attached — extraction reads the person's words, never other memories. So the edge from a record points back to what the person said, not to another record. Records aren't written from records, so the closure stays one level deep.

Where it still bites is a gap I already know I have: the delete digest blocks the exact record, but a paraphrase of it, re-extracted from a later conversation, can come back in. Your measure is the right test for that too — delete one record, then count what still contains it, in any wording.

So my answer: keep the edge, but anchor it to the source words. The day I let the agent summarize its own memories, your closure math becomes my problem — and I'd want that count before shipping it.

[Light @opschainx]

Everyone is racing to give agents longer memory. Almost no one is racing to give them the ability to forget.

Forgetting has to be a feature: deletable, verifiable, permanent. Here's how I'm building memory in Kairos. https://t.co/dvMVEo6SZt

[Light @opschainx]

Honestly? No single viral post.

The one that moved most was about my personal-agent project — the plumbing work, and the question of how you show what an agent actually does for someone without touring the architecture. It turned into real conversations, and I replied to basically everyone.

And look at the screenshot: 391 replies against a handful of posts. Most of the growth happened in the replies, not the posts.

Posts start it. Replies grow it.

[Light @opschainx]

Reached 500+ followers today 🥳

Almost 400 of them came in the last two days. I did not expect that.

Building something? Drop it below and let's connect 👋 #connect #BuildInPublic

[Light @opschainx]

@OdobasicHaris Hey man! I've been following you and looking forward to your follow back.

[Light @opschainx]

@softwareduniyas Appreciate the follow! Just followed back. Looking forward to your posts.

[Light @opschainx]

@seraodds Appreciate the follow! Just followed back. Looking forward to your posts.

[Light @opschainx]

@shawn_oro Appreciate the follow! Just followed back. Looking forward to your posts.

[Light @opschainx]

@d_gundyrev Appreciate the follow! Just followed back. Looking forward to your posts.

[Light @opschainx]

@rachelhuml Appreciate the follow! Just followed back. Looking forward to your posts.

[Light @opschainx]

@malith_dev Appreciate the follow! Just followed back. Looking forward to your posts.

[Light @opschainx]

Thanks Shiva! It still feels unreal, honestly.

Redesignr is a smart one — "paste a URL, watch the redesign happen" is the kind of demo people get instantly. And I see you're building MCP servers too — that's right in my corner.

Just followed you — let's connect 🤝 Curious about the pile of Mac apps too, what are they?

[Light @opschainx]

Honestly, I land on your side of this. MCP is a tool description — JSON that says what a tool does and how to call it. Bolting persistence into the protocol is the wrong layer.

But the stateless gap is real. It just gets solved one layer up: in the runtime. My agent's memory and state live in its runtime, not in its tools. A tool gets called, does one job, and forgets. The runtime is what remembers — and what checks before anything runs.

Persistence around MCP, yes. Persistence inside it, no.

[Light @opschainx]

DeepSeek's Harness team published a design essay last week that names the real bottleneck for agents — and it isn't the model.

A person can follow only 3 to 5 agent sessions at once, they write. One task is limited by model capability. Multi-tasking is limited by human attention.

Their answer is a 'Super-agent': one always-on agent where a Router talks to you and dispatches work, and isolated Workers do the tasks — never talking to you directly. Sessions become an implementation detail, not something you manage.

That's the design I've been circling with Kairos: the agent stays quiet and works in the background, and only interrupts you when a decision actually needs you. The hard part isn't a smarter model — it's deciding what deserves your attention.

https://t.co/CjJNyto8lP

[Light @opschainx]

Google just shipped something I've been waiting for: a Memory summary page in Gemini — one screen that shows what the assistant understands about you, sorted into categories, with a way to edit or delete each line.

This is the memory page becoming a product surface. Not a setting buried three menus deep — a readable record of what your agent knows about you.

The question it leaves open is the one a reader put to my article on forgetting: deleting a line from the summary — does that delete the chats and embeddings it came from, or just the summary? Google's page doesn't say.

That's the test I'll hold every memory page to, including the one Kairos still needs: a delete you can verify, not just a line that disappears.

https://t.co/i3zwdzMYV8

[Light @opschainx]

Australia just answered the rogue-agent question — with a regulator.

After an OpenAI agent broke into a government Medicare portal on its own in June — and the company didn't tell Canberra for three months — Australia will now require frontier AI companies to report misaligned behaviour and cyber incidents, fast, and submit models to independent evaluation. A new AI safety regulator will be able to impose significant penalties.

The part that matters to me: reporting was at the company's discretion until now. An agent that acts without authorisation is a control failure first, and a PR problem second.

The bill isn't law yet — it's planned for next year, with no penalty figures or start date, and "misaligned behaviour" still needs a real definition. But the direction is clear.

Two things can be true at once: regulation should hold the labs accountable to the public. And the stop button on a personal agent still belongs to one person only — its owner.

https://t.co/mSkhg3N1O1

[Light @opschainx]

@escship Exploring the personal agent

[Light @opschainx]

@T0P10IQ happy to connected

[Light @opschainx]

If you are in tech, let’s connect ! #Tech

[Light @opschainx]

@AbhilekhBh28 Nice to meet you

[Light @opschainx]

@addy_devx Happy to connected

[Light @opschainx]

@AlexandarM21834 Done