The shared-context tax is the real bottleneck in your AI stack
Every cross-border operator I know has quietly become an AI power user. Your sourcing lead runs ChatGPT to rewrite supplier emails. Your ads person lives in Claude for creative variants. Your VA in Manila drafts listings in Gemini. Individually, everyone got faster. Collectively, the team got dumber — because none of that context is shared. The product research your ads person did last Tuesday is trapped in a private chat thread. The supplier negotiation notes your sourcing lead generated never reach the person writing the Amazon bullet points.
That’s the problem Spaces is trying to solve, and it’s the first AI tooling pitch in a while that maps cleanly onto how cross-border teams actually work. Let me explain why this matters more than another prompt library, and where I think the pitch oversells.
What Spaces actually is, and why the “shared space” framing is the right primitive
The team behind Spaces — launched by John on Product Hunt — built a desktop app where each project gets one shared space. Inside that space, humans and their AI agents work together: shared chats, shared files, and “specialists” (a researcher, a copywriter, a launch lead) each carrying its own instructions, memory, and tools. Anyone in the space can put any specialist to work. Routines handle standing work — a morning brief, a Friday recap — and only ping a human when a decision is needed.
Two architectural choices stand out. First, you bring your own LLM provider — ChatGPT, Claude, Gemini, or even local models running on your own machine. Spaces isn’t reselling tokens and there’s no new subscription for inference; you’re buying the space, not the tokens. Second, your agents, API keys, and connected accounts stay on each person’s own computer. The cloud syncs only the space itself — chats, files, routines — nothing else.
The desktop app is free to work solo, no account required. A shared space runs $10.99/year per person.
Why the “one space per project” model beats “one workspace per company”
Most AI collaboration tools I’ve tested — Notion AI, Slack AI, the various Microsoft Copilot integrations — organize around the organization. That’s backwards for cross-border teams, because your work doesn’t organize that way. You don’t have “a company context.” You have a Q3 Amazon launch, a TikTok Shop test, a Temu listing refresh, and a supplier audit running in parallel, each with different people, different agents, and different memory requirements.
A sourcing agent that remembers MOQ negotiation history for your Shenzhen supplier is useless inside a space where your TikTok creative team is iterating on hooks. Spaces gets this right by scoping memory and tools to the project, not the org chart. That’s a small architectural decision with large operational consequences.
How it differs from what you’re probably using today
Let me be specific about the comparison set, because “AI workspace” is a mushy category.
Versus raw ChatGPT Team or Claude for Work: These give you shared billing and admin controls, but the chat histories are still fundamentally individual. You can share a link to a conversation, but you can’t build a persistent shared memory that a specialist agent draws on across sessions. The context lives in people’s heads and browser tabs.
Versus Notion AI or Coda AI: These are document-first. The AI is bolted onto a wiki, which means the primary artifact is a page, not a conversation. For research synthesis that’s fine. For iterative agent work — “run this supplier email through the negotiation specialist, then have the copywriter turn the outcome into a listing angle” — it’s clunky.
Versus Zapier or Make automations: These are trigger-action pipes. Great for “when a Shopify order comes in, log it to a sheet.” Terrible for anything requiring judgment, memory, or a human in the loop. Spaces’ “routines” concept sits closer to this end of the spectrum, but with an LLM doing the work rather than a deterministic workflow.
Versus building your own stack with OpenAI’s Assistants API or LangChain: This is what the more technical operators among you have already done. You get full control, but you also get to maintain it, and you get to explain to your ops manager why the agent broke at 2am during a flash sale.
The honest positioning: Spaces is trying to be the “shared context layer” that sits above whichever model you prefer, without becoming a model reseller or a workflow engine. That’s a defensible wedge.
Why Amazon sellers should care more than Shopify ones
Here’s my judgment call. If you’re a Shopify DTC operator, your team is probably 3–8 people, most of your context lives in Klaviyo flows and a Notion doc, and the pain of fragmented AI chats is annoying but survivable.
If you’re an Amazon FBA brand owner, the pain is acute. Your operation spans Amazon Seller Central, Helium 10 or Jungle Scout for research, Keepa for price history, a supplier relationship in Shenzhen, a freight forwarder, a TikTok Shop account you’re testing, and a Temu listing you’re defending against. Every one of those surfaces generates context that some other part of the team needs. The cost of NOT sharing that context — misaligned listing copy, duplicate supplier outreach, ads targeting a keyword your sourcing lead already knows is saturated — compounds weekly.
A shared space per SKU launch, or per marketplace, is a more natural fit for that operation than for a lean Shopify brand.
What cross-border sellers can actually borrow from this
Even if you don’t install Spaces, the design principles are worth stealing for your own stack. I’ve been running variants of these for the past six months and they’ve moved the needle more than any single tool purchase.
1. Scope memory to the project, not the person. When you brief a freelancer or a VA, don’t just hand them a Google Doc. Hand them a folder with the supplier emails, the competitor screenshots, the keyword research export, and a short “here’s what we’ve already tried” note. Then tell them to keep their AI chats inside that folder’s context. The difference in output quality is not subtle.
2. Separate “specialists” from “generalists” in your prompting. A single mega-prompt that tries to be a researcher, a copywriter, and a media buyer produces mush. Three separate system prompts — each with narrow instructions and its own reference files — produce work you can actually ship. Spaces productizes this; you can replicate it today with custom GPTs or Claude Projects.
3. Treat “routines” as your ops calendar. The morning brief and Friday recap patterns Spaces describes are just structured recurring prompts. You can run these manually in 15 minutes a week and get 80% of the value: a Monday prompt that summarizes last week’s Shopify and Amazon Seller Central numbers, flags anomalies, and drafts three actions.
4. Keep your API keys and connected accounts local. This is the underrated part of the Spaces pitch. If you’re plugging AI into your Shopify Admin API or your Amazon SP-API credentials, the last thing you want is those keys sitting in a third-party cloud you don’t control. Local-first is the right default for anything touching live seller accounts.
5. Bring your own model, deliberately. The “use any LLM provider” line matters more than it sounds. Different tasks warrant different models. Long-context supplier contract review? Claude. Numeric analysis of ad spend? GPT-4 class. Bulk translation of listing copy into five EU languages? A cheaper model or a local Llama instance. Locking into one vendor is a tax you don’t need to pay in 2025.
Where the math breaks
The pricing is aggressive — $10.99/year per person is essentially free at the team level. But the real cost isn’t the subscription, it’s the token spend, and Spaces explicitly doesn’t cover that. If your team of six each burns $40/month in API calls across providers, your “cheap” shared workspace is actually a $2,900/year line item once you add inference. That’s still probably worth it if it kills duplicate work, but don’t let the $10.99 anchor fool you into thinking this is a rounding error.
There’s also the migration cost. Getting a team to move their AI habits out of personal ChatGPT tabs and into a shared desktop app is a behavior change project, not a software install. Budget for it.
Where my judgment says Spaces falls short
I haven’t used the product hands-on, so treat this as informed skepticism rather than a verdict. But three things concern me from the pitch alone.
Desktop-only is a real constraint for cross-border teams. Your VA in the Philippines, your sourcing agent in Shenzhen, your 3PL contact in Poland — they’re not all going to install a desktop app, and some of them are working from machines you don’t control. A web client is table stakes for distributed teams. The pitch says “desktop app” without qualification, so I’ll assume web isn’t there yet.
“Local-first” and “shared space” are in tension. If agents, API keys, and connected accounts stay on each person’s own computer, then what exactly is shared? The chats, files, and routines sync — but do the agents sync with their memory intact? If my copywriter specialist has been trained on three months of my brand voice notes, and my new hire joins the space, do they get that specialist or a blank one? The pitch is fuzzy here, and it’s the single most important question for anyone evaluating this.
The “specialist” concept is only as good as your prompt discipline. Every tool that promises “agents with their own instructions and tools” eventually runs into the same wall: most operators write bad instructions. There’s no mention of templates, onboarding, or a specialist library. If Spaces ships with a decent set of pre-built specialists for common e-commerce workflows — supplier negotiation, listing optimization, ad copy testing — that changes the adoption curve dramatically. If it’s a blank canvas, you’re paying for a framework and supplying all the value yourself.
No mention of compliance or data residency. For sellers handling EU customer data under GDPR, or dealing with DSA obligations, “the cloud syncs only the space” is not a sufficient answer. Where are those servers? Who’s the data processor? Not disclosed in the launch post.
The competitive risk nobody’s talking about
OpenAI, Anthropic, and Google are all racing toward team-tier products with shared memory and agent orchestration. Microsoft is bundling it into Teams. The window for a neutral, provider-agnostic layer is real but not infinite. Spaces’ best defense is exactly the thing it’s pitching: neutrality. If it stays genuinely model-agnostic and local-first, it can survive the platform squeeze. If it drifts toward favoring one provider, it dies.
What I’d watch / test next
Three concrete things to do this week, whether or not you touch Spaces.
First, audit your team’s AI context leakage. Ask each person on your team to list the three most valuable AI conversations they’ve had in the past month. Then ask: did anyone else on the team benefit from that output? If the answer is usually no, you have a shared-context problem, and it’s costing you more than any SaaS subscription.
Second, stand up a manual version of one “space” this week. Pick your next product launch. Create a shared Google Drive folder, drop in every relevant doc, and require everyone to run their AI chats inside that folder’s context — even if it’s just by pasting a shared “context block” at the top of every prompt. Measure whether the quality of outputs improves over two weeks. That’s your baseline before you pay for anything.
Third, if you’re technical, price out the DIY version. A Notion database plus OpenAI’s Assistants API plus a shared Google Drive gets you maybe 60% of what Spaces promises, for the cost of your time. If that 60% is enough, skip the tool. If the remaining 40% — the local-first security model, the specialist abstraction, the routines — is what you actually need, then $10.99/year per person is a cheap experiment.
What I’m watching for next: a web client, a specialist template library, and a clear answer on how agent memory behaves when a new person joins a space. Ship those three things and Spaces becomes a default recommendation for cross-border teams. Ship none of them and it stays a clever desktop app for solo operators who already had their workflow figured out.
