Tom Goodwin

I do find this just amazing

The fact that little kids put everything in their mouth immediately doesn’t seem great for evolution

@TheManxomeFoe90 Yes but expecting it to all come together is.

@tomfgoodwin Agents executing pre built software operations/tasks by taking user text input and translating it into parameters to send to API calls isn’t magical thinking or flawed.

The thinking that because LLM's can do tasks, not jobs, that the solution is AGENTS, because agents can do many tasks, is very flawed. It's sort of magical thinking, blockchain fixes this, DAO's will sort this, it will be resolved by Web3, it's just not how the world works

@i_am_anmolg What have you been doing with them?

@tomfgoodwin I've been using agents under strict supervision and I must say, these agents are nothing like before. The way world works is already changing and will keep change rapidly.

It was always interesting to me 50% of Clients would see an ad campaign not hit their targets, so spend more. & 50% would see it failing and spent less.

@calebkramer This sounds desperately like a company trying to use AI in places it really doesn't fit. To solve problems that really, really, really don't exist.

Starbucks is enhancing its human edge, using AI to let baristas do what they do best. But there's going to be a fine line between this and RentAHuman.

@satyanadella That Cafe poster is criminally bad. Is it designed by the team that gave us liquid glass?

Great to see our new image model from our Superintelligence team rolling out in Copilot and coming soon to Foundry for enterprise customers.

Feels more black mirror than black mirror Someone recording themselves cleaning your house to train a robot

@shahjahan_ch Maybe. Are you just gonna get ai to scan my life and write it? ;)

@tomfgoodwin Hi @tomfgoodwin - can we profile you?

I want to live in a world with almost no productivity hacks and lots of guidance on how to enjoy your time or life, or find passions. ( I don't need these tips, but I feel like many do )

@aralin_eth superb answer, thank you.

Agents are basically a synonym for memory context window. We assign model and skills and some other context to agent and with that base context we send him to do a series of tasks. Now if something else needs to be done, we switch context, same as in any other multi-tasking. It would be more costly to try to build the context from scratch every time and it would lead to errors if irrelevant stuff like skills and conversation history was present in agent’s memory as it sends the whole thing to LLM model with additional prompt. So we separate into multiple context windows that we store separately and call it agents just to anthropomorphize it.

I’m surely being stupid. But if AI is rather unconstrained by expertise or capacity or to some extent speed Why do we need to divide tasks or departments to 9 agents ( the marketing agent, the optimization agent etc ) to each do one thing. And then another agent to manage the swarm. Cant one agent just be doing it all you know. It seems very skeuomorphic. Will we have HR agents to make sure the agent agents are being looked after ? A office canteen manager agent to feed the agents ? Seems daft

They only exist because it's a mega mega cheap fast way to do thought leadership, the fact it's useless isn't the impediment one would expect.

I do think most Surveys asking companies / CEOs / CDO's, how they use AI, or what they plan to spend money on, or what they are thinking about , or what themes they see, etc etc Are mostly a waste of time. People just say what they think they are supposed to. Nobody wants to be boring. Nobody wants to see ignorant. We get a circle of fluff 93% of people are excited by agentic workflows. 91% of CEO's think Quantum is a big deal. 81% of CDO's think I9P5 protocol is interesting, even though we made it up.

@kaseyklimes But but but but it's an AI powered rube goldberg machine, It can be even more messy. , It can be even more pointless. , It can be even more inelegant. , But just look at how clever it is.

we've already been through a few bundlings and unbundlings here: 1. hit the limitations of current models 2. break out responsibility to individual agents 3. the models improve, rendering your specialized division of labor irrelevant 4. realize you built a rube goldberg machine, strip it down to a single elegant agent 5. push your system to handle more complexity at higher levels of abstraction return to step 1

It would be interesting to see some experiments where people try certain workflows and compare the results. I don't think we should take it for granted that more agency equals better. Seems quite likely that we're creating things where errors can compound, And where brilliance can be made grey. That's not to say it isn't a better way to work, but it may not be, net net.

@tomfgoodwin Models are fine-tuned during training + learn in context, so there will always be benefits to specialization (or trade offs when avoiding it). Note that some models handle all this internally: the reason with themselves from different perspectives.

@charlesarthur @Sonos But software is easy now. , Code writes code. yawn LOL

Amazing. One of the things people complained about with the updated @sonos app was not being able to edit alarms. Two years on: still can't edit alarms. You can change stuff, but it isn't saved. Just astonishing.

My dog would go absolutely mental at every vacuum ever, robot ones the worst. Then we got a Dyson and it's fine. Am I alone? Is there an ad campaign in this?

Do you think we'll ever get ads as good as this again? Is there a reason why? youtube.com/watch?v=-M3AjO…

@aadhicash @nejatian I've never heard a wonderful lifestyle described as a distraction before.

@nejatian Is the scene really that great?? Feels like there are a ton of distractions compared to SF.

Miami is such an underrated city for builders. If you are a world class engineer and considering moving here, I’d love to tell you about the benefits of Miami and Florida (and Opendoor!)

Time and time again , and increasingly so, the constraint isn't can you build it, but the quality of the idea. If thoughtful types sat in this room making pages of ideas, they could then hand this off to the Hackathon, and select things to jam on. "What if Apple Car Kit was all it could be" "What if you could own a fraction of" "What items would AI really improve and how" "What dreams are unrealized" "What if a retailer could xxxxx" "What software would Dyson make" etc etc

Someone should invent a new type of hackathon that's sort of the opposite- its called a "What If" Rather than tech folk play around around with great tech to see what they can make We get imaginative types in a room & just share solutions, or ideas, or things that should exist.