We tried to model some of that here where the harness has, of course, these core facts. It also has these dated logs of kind of your past conversation history centrally. And then also these kind of scratch pad notes … little things that you've been working on right now that are designed to be ephemeral, and they will go away, but they might be relevant just for the next 30 minutes or an hour.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson has just compared the agent's memory to how a brain keeps relevant facts in mind, and is describing the three memory layers in the Grok Bot harness.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 33:51.

Checked against the source [ Grade B ] Said in the last year Sells AI tools

Filed under

Who is speaking

Lee Robinson, Machine Learning Engineer, Model Behavior, Cursor. Works on the team that trains the Cursor models he describes, so he is close to the training process; Cursor sells AI coding tools and its own models, and the talk is promotional of its approach.

SpaceX (SpaceXAI) sells AI or agent tools or services, so it has a reason to publish flattering numbers about AI. The passage stays in the archive; read it with that in mind.

How this was checked

Checked against the source by a first reader, then confirmed independently by a second reader who fetched the source without seeing the first reader's notes (Independent review agent (Claude Sonnet), blind second pass from passage text and citation only, no repository access, 2026-11-07).

Whisper transcript, cue at 34:04. 'We tried to model some of that here where the harness has, of course, these core facts. It also has these dated logs… scratch pad notes… for the next 30 minutes or an hour.' The ellipsis covers 'just like'. The transcript spells it 'scratch pad', as the passage does. The cue starts at 34:04 and the stated locator is 33:51, which is a small mismatch in the locator only. The wording is confirmed.

How to cite this passage

Lee Robinson (SpaceX (SpaceXAI)), “Stanford CS146S: How always-on agents work”, leerob (YouTube), 2026-10-11, at 33:51. https://www.youtube.com/watch?v=L1OBRtfPS34 Archived at https://the-ai-record.com/passages/spacex-grok-bot-memory-core-facts-logs-scratchpad-2026

More from the same source

It's increasingly common now at SpaceX where we see bots joining meetings for employees … but they want their bot to go and just take notes and come back to them, which has gotten funny.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson is listing the events that wake an always-on agent, such as a Slack message or a meeting. The cut skips 'where they could make it to the meeting', which both passes transcribed the same way but which reads as a mis-hearing.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 11:26.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

One of the principles that we had at Cursor and now at SpaceX is to delete the product. And what we mean by this is you kind of have to internalize and assume that the models today are the worst they'll ever be, which means that in six months you might need to completely rebuild the UI.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson is describing the product principle his teams followed at Cursor and now at SpaceX while building harnesses for always-on agents.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 37:45.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

So I talked about how the agent has this computer, this Linux box. This is running a firecracker virtual machine, and the reason this is really helpful is because it's very easy then to essentially snapshot the state of the machine and shut it down and start it back up.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson is describing the infrastructure behind Grok Bot, the always-on agent his team at SpaceX builds, and the computer each agent gets in the cloud.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 12:24.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

One trick that we do, which I think is really helpful, is while you're still within the warm cache period, you should probably do the compaction right now, because you have this long conversation … going to be much cheaper. So just go ahead, do the compaction, and now when the user comes back at 30 minutes or whatever, they're starting from this summary, and the summary now is going to be much cheaper for them to continue adding messages on.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson is describing how his team keeps the cost of long agent conversations down by summarizing the context before the model provider's prompt cache expires.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 32:20.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

… there's only one tool that communicates between the client and the server. And it's just send user. And the great thing about this is that it's actually very architecturally simple where the client is just texting the server in some ways.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson is describing how the Grok Bot harness differs from other tools on the market. The quote starts mid-sentence, after a lead-in the two transcription passes heard differently.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 18:21.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

So we use an open source tool called temporal, which is interesting to look into if you're curious, so that we don't have to rebuild a lot of this very complicated stuff from scratch.

Lee RobinsonWorks on Grok Bot at SpaceX (formerly Cursor) · Practitioner, SpaceX (SpaceXAI)

Context. Robinson has just explained that an agent task that fails midway should resume at the failed step, not replay earlier steps, and says Grok Bot uses Temporal for these durable workflows.

Talk: Stanford CS146S: How always-on agents work, leerob (YouTube), October 11, 2026, at 17:08.

Checked against the source [ Grade B ] Said in the last year Sells AI tools Passage page

Something wrong? Report a correction. JSON: /api/v1/passages/spacex-grok-bot-memory-core-facts-logs-scratchpad-2026.json