←
Everett11,408 posts
Everett avatar

Everett

@retttx

GTM Engineer, Ant, Dad x2

◷ Joined May 2012▧ 626 Photos and videos
Matt Ambrogi avatar
⟳ Everett retweeted
Matt Ambrogi@matt_ambrogi·Nov 17

Lot of people asked about this. You can basically make your MCP self improving by doing this: - Add a submit_feedback tool to your MCP - Agent using your MCP will write detailed issue including proposed fix. - Have it send to an internal Slack channel - Have Claude in Slack channel and tag it to put up a PR when you agree w proposed change.

Post attachment

This is why you should actively invest in an embedded agent and an MCP. MCP observability only allows you to see tools called and you have to reverse engineer intent. If you have an actively used embedded agent, you would have seen this query and been able to ID that the proper filters were not in place. Fix the tool and it benefits both surfaces. You can also get a lot of benefits by adding a submit_feedback tool to your MCP to gather this data.

Everett avatar
Everett@retttx·22h

The good news is we can eliminate measles again - a second historic achievement

This is pitiful. Eliminating measles was a modern miracle. Now we've seen more cases in the last 18 months than in the previous 25 years combined, because a miraculous and easily available vaccine has been needlessly polarized. Spurning progress is a choice.

Fernando Borretti avatar
⟳ Everett retweeted
Fernando Borretti@zetalyrae·Nov 15

Crazy that Kurzweil's "singularity in 2045" sounds delusionally conservative now.

Everett avatar
Everett@retttx·Nov 14

Stuck in Jersey after the flight situation. I mean I live here, but I’m stuck here too.

Everett avatar
Everett@retttx·Nov 13

what is my usage going to be? idk it's ~your~ usage!! i am merely providing the means to use it

brendan short avatar
⟳ Everett retweeted
brendan short@brendanjshort·Nov 13

Which sales jobs will agents take? Dan Lee (@_dan_lee), Nooks CEO, says: "I think it goes back to what is an easy persuasion problem and what is a hard persuasion problem. An easy persuasion problem is a transactional deal where they mostly want to buy already, and they’re on your website, and you just need to answer a couple questions to get them over the line. Whereas a hard persuasion problem is: you’re in an enterprise deal cycle, there are 20 people involved, half of them have never heard about your product and are not prioritizing the problems you solve, you’re about to enter a budget freeze. And the easy one... can be fully automated. Agents are going to be better than people at that, and it’s going to be a better buying experience too. But on the other end... you need to build relationships and trust and really deeply understand their problems. You’re going to have a 10x rep who’s augmented with agents, where the agents are helping them focus all their attention on deal strategy, on their human relationship building, on working more deals like this, and agents are making sure nothing slips through the cracks."

1:25
Aja Frost avatar
⟳ Everett retweeted
Aja Frost@ajavuu·Nov 11

this is bigger than people realize. up till now building for MCP has been kinda scary 😟 because you're giving your data to the LLMs and your users will attribute some — if not most — of the value to the frontier model, not your product. you do it anyways because you can't afford not to. but incorporate your own UI, and now the user remembers who you are and credits you more for the experience.

Also new: with interactive connectors, you can now open the full version of a tool inside Claude. Work in Adobe, Atlassian, and Box beside your conversation. Make changes yourself, or ask Claude to make them with you. Rolling out on web and desktop.

1:00
Paul Klein IV avatar
⟳ Everett retweeted
Paul Klein IV@pk_iv·Nov 9

fable 5.2 has some of the best computer-use capabilities we've seen

Post attachment

Fable 5.2 scores 92% on HardBench, our curated benchmark of difficult computer-use tasks. It beats the previous champion, Opus 5.5, in accuracy, but costs 2.5x more and is 13.76% slower on average.

Post attachment
Everett avatar
Everett@retttx·Nov 9

You can probably best see this in the huge jump in "visual chart recognition" at the bottom of these benchmark. That kind of capability makes a big difference in Fable understanding what is on the screen / dash and making the correct next decision.

Post attachment
Everett avatar
Everett@retttx·Nov 9

Fable 5.2 is excellent at computer use, far better than Opus and what the benchmarks show. In GTM you should use it to control all your UI based tools across multiple tabs and expect to be able to solve systems issues that occur across tools. For example, you might get a slack report of a routing issue that starts in LeanData, but actually relies on data from Clay, the data warehouse, and your CRM. You can reliably expect 5.2 to dig into every surface you can log into and update configs accordingly - even if certain settings or views are not available via MCP or API. Make sure you have 'desktop browser' set in Claude or kick things off via the chrome extension.

Alex Albert avatar
⟳ Everett retweeted
Alex Albert@alexalbert__·Nov 9

btw robotics will fall to the bitter lesson too

Fable 5.2 shows substantially improved skills in controlling physical robots, even on tasks it has never explicitly been trained to do. On RoboDojo, a benchmark with a simulated two-arm robot, it outperformed other models and most AI systems built to control robots.

0:23
Tom Reed avatar
⟳ Everett retweeted
Tom Reed@mentalgeorge·Nov 5

Still the funniest joke I've ever heard on an AI podcast

Post attachment
Everett avatar
Everett@retttx·Oct 30

You never think you're going to need that 2 TiB hard drive.. until you do