From Amazon's perspective, that would be very much the wrong way.

Amazon's concerns here aren't technical. It's entirely about being disintermediated.

https://marketplace.openship.org

It’s an interoperable marketplace that uses MCP-UI to connect to seller websites directly. It brings the storefront and cart directly into the chat.

I'm using Claude for everything from finances to dinner planning to project planning. If Amazon doesn't let Claude search and buy, I'm not abandoning Claude, I'm abandoning Amazon.

As long as my actions are as simple as possible, I don't care whether Claude is searching 5000 online retailers or only 4999, minus Amazon. I'm certainly not going to say "ok, this project requires buying 13 things for about $500, but I don't see amazon in the list, so let me go manually search for each one on Amazon and see if they might have a better price or something."

This is Walmart / Apple Pay all over again. Yes, near-monopolies, lots of power. But they got there by making things easy for consumers. Making things difficult to force consumer behavior never ends well.

Half the time I can't even find the category page. And if I do find it, the filters will just be useless anyway.

Amazon is one example. If my basement AI drives my browser and never clicks an ad, Amazon loses big.

Or, the ad industry in general -- imagine the basement AI filtering out all ads at the analog level so nobody in my house ever sees an ad, and no software update can change that.

Or other optimizations. Right now companies enjoy a power imbalance because they have people who ruthlessly optimize 24/7 to save a buck. I can't do the same, because I have to work, and also prefer to spend my personal time with family -- a foolish thing to do in a capitalist society, at least from a capital perspective. Imagine I have an AI wheeling and dealing on my behalf 24/7 too. Especially in an age of online shopping and product deliveries, my AI will absolutely order from your competitor to save a penny, my AI will absolutely spend 400 hours disputing a medical bill I don't think is right, barely an inconvenience to me.

Agents that scrape and digest web data are entirely anathema to brand-reliant and advertising-reliant vendors and marketplaces. Even human assistants could be trusted to develop brand attachment and respond to advertising meant to shape their preferences. And while LLM agents are still manipulable, that's a frontier that nobody really knows how to navigate effectively yet, at least not to the degree of optimization modern vendors and marketplaces (especially Amazon) have acheived.

It's quaint to imagine what AI agents might do for consumers in the naive internet that preceded their emergence (if you ignore their vulnerabilities and flaws), but that's not what they'll be able to achieve as that same internet responds to them -- at least not at Meta scale. It's not very different than how social media companies once offered, and then aggressively withdrew and policed, API's to their graphs and feeds. Those open API's simply aren't compatible with the business objectives that keep these companies running, no matter what cool and profoundly user-friendly things could be built on them.

The same kind of incompatibility applies to AI agents when it comes to e-commerce. It won't be a permitted thing until, and except for where, vendors/marketplaces can assert control over the agents themselves and manipulate them to ends comparable to what they can achieve with human consumer sentiment already.

I think the arguments for these kinds of near-monopoly companies being true market-driven capitalism seem more and more false.

The idea that Amazon will tell you that you don't even own the right to your own shopping experience tells you everything you need to know. Apple owns your device and Amazon owns the right to show you whatever price/product they want, regarless of your search intent.

- Dusted off a 3d printer that wasn't being used enough, largely because of the tax of setting up profiles for materials and each print. That was fun once, I'm too busy now. Was able to get claude to use MCP and configure the slicer just right for the job and filaments.

- Needed to get IPTV/etc set up for football season. If you've done this you know what a pain it is, how EPG's don't always work right, finding the 50 channels you care about out of 80k channels is a nightmare, etc. Pointed claude at my utility server, it got Dispatcharr configured, wrote scripts to parse channel guides from sources not covered in the EPG, scheduled to update daily, deployed.

- I'm considering getting back to 3D printed clay, which is fun but challenging. Based on a thread here on HN, had claude go investigate using FPGAs (which it would program) for a realtime flow modulation system that would remove the variability from slight changes in viscosity within a batch of clay. Not sure I'll pull the trigger but I have a much deeper understanding of the solution.

- Realized I was paying like $30/mo to RocketMoney and Copilot (finance, not MS), for rare usage and basically just occasional "find subscriptions I forgot about" usage. Got claude connected to plaid via MCP, found more than those other two ever did, got it to set up cron to send me email weekly with current subscriptions and any price changes.

That's literally in the past 4-5 days. I find great value there, YMMV of course.

As a baseline which many skeptics will have a hard time accepting for their own usage - I think there is a correlation between utility and trust (even which is not truly earned). If you don't let it capture memories and learn about you, and you don't connect it to external tools like email, it's less useful. But I think of it usually like a ChatGPT w/ extra tools and knowledge.

I just moved, and last night I asked my agent how to test how safe the soil in my (urban) garden was, and it found some warnings from my city government, and suggested a particular test service that is available in my area. Since it knows exactly where I live, it was able to find recent warnings of contamination in my neighborhood, and suggest specific chemicals to test for.

Additionally, I give it (read only) access to my email, and it'll ~2x a day notify me of some important email. I'm working with a contractor, and getting emails from them bumped, or getting emails about AWS usage alerts pulled out of inbox and notify me directly is super helpful for filtering the noise of my crazy email volume of 99% spam.

Also, as others have said, it's easy to take a picture of a flier, and say "add to my calendar".

At work? I've got ADHD, so the idea of making a to-do list is fine but keeping it up to date was a challenge. I've created a set of agentic skills that comes through the last 36 hours of emails, meetings, chats, and gathers up all the work I volunteered to do (or was told to do), triages them across urgency/impact, and sorts them into Do/Defer/Schedule/Delegate piles.

Last, I created an app to give a UI to this data to see and act on it all, including a way to flag a to-do that was found and enter why I thought the status or categorization was wrong. On the next run, a review skill reads those comments, makes changes to the other skills accordingly, and sends me an email with details of the changes, and the commit I can use to roll back the change if I'd like.

I've been running it since April and while there were a lot of updates at first, I haven't had to flag anything for review in months, and I get to feel a whole lot better about managing my to-do pile.

I'm looking at open-sourcing it, but since I created it with work resources on work time, I need to jump through a few hoops first.

1) Staying on top of my kid's school obligations, homework, etc. School Learning Management Systems are a hot mess, and I use agents to consolidate and collate, to keep track of the homework that has been done, what needs to be done, and even pre-grade homework to give feedback (most of the work is done on digital platforms too). It's been immensely time saving for these kinds of things. 2) Multiple instances of watching things passively: hotel room prices, availability of tickets, finding the best place to sell things online. In general things that require a lot of browsing, collating, and watching. These are the kinds of things that specialized software products could do as well, but with an agent they are just a prompt away, and you get exactly what you want. (More or less.)

- travel research for places where I need an apartment with requirements one cannot just filter for

- gardening tips based on photos of my plants

- financial planning (similar tips to what I would get for more money, without paying per hour until I know much more)

- take photo of all physical spam and have me removed from the list

and many more…

If the access to this product went away I will survive, but it will be annoying and waste my time.