ResearchOctober 3, 2026 · 5 min read

I searched Reddit for this problem. What I found was everyone.

Before writing a line of code I searched Reddit, Hacker News, Twitter, and DEV Community for anyone complaining about switching between AI tools. I expected a few posts. I found hundreds. Here's what people actually said.

I have a rule I follow before building anything: search for people already complaining about the problem. Not to validate the idea — I already knew the problem existed because I had it myself. But to understand the shape of it. How people describe it. What language they use. What made them frustrated enough to post about it publicly.

So before I wrote a single line of Crossagent, I spent a few hours on Reddit, HN, Twitter, and DEV Community searching for versions of this question: “How do I use multiple AIs at once?” and “I keep switching between Claude and ChatGPT.”

What I found was that this problem has a name people keep independently inventing: human middleware. The person manually carrying context between AI systems that refuse to talk to each other.

The LinkedIn post that stopped me

The clearest statement of the problem I found wasn't on Reddit or HN. It was on LinkedIn:

“

I hate switching between ChatGPT and Claude because with every new chat I have to re-explain my project, files, and goals. It's a total waste of time.

Manthan Patel·LinkedIn ↗

That's it. That's the entire problem in two sentences. The tabs aren't the issue. The re-explaining is the issue.

Hacker News: the workarounds people built

On HN there was an Ask thread: “How do you manage context/memory across multiple AI tools?” The original poster described exactly what I was doing:

“

I use Claude for some tasks, Cursor for coding, ChatGPT for research, and Perplexity for quick lookups — but the problem is none of them know what I've discussed with the others, so I end up re-explaining the same context repeatedly or copy-pasting from my Notion docs.

Ask HN thread·Hacker News ↗

One reply that stood out: someone wrote that they keep three terminal sessions open and tell Claude and Codex to maintain a shared markdown file with progress, requirements, and decisions. That is an extraordinary amount of work to solve what should be a solved problem.

Another HN thread — “Is AI context-switching exhausting?” — had a reply that described the psychological toll in a way I hadn't expected:

“

It's like gambling — you're chasing a result that feels close but never actually arrives. That's where the exhaustion comes from.

andy99·Hacker News ↗

DEV Community: developers describing it precisely

“

I spent a week tabbing between four AI chats to answer one question — ChatGPT for summaries, Claude for code, Gemini for its larger context window, and a local model for sensitive data. Every time I switched, I had to re-explain the project: what had already been tried, which ideas were rejected, what decisions were made.

thatcooperguy·DEV Community ↗
“

I use Claude, Gemini, ChatGPT, Manus and Cursor, sometimes for the same aspect of the same project. As a result, I can end up spending a lot of time moving outputs between tools... This gets annoying.

lbdev22·DEV Community ↗

The irony post on Substack

This one made me laugh:

“

I was building in public about 'focus' while context-switching between 6 AI tools.

KP·Substack — Build In Public Hub ↗

The people most aware of the problem are often the people posting productivity content about AI. They know the workflow is broken. They just don't have a fix.

Then I found the Workday study

The most authoritative framing of the problem I found wasn't a Reddit comment — it was a May 2026 research report from Workday. Their VP of something called it directly:

“

Too many employees are serving as the human middleware between disconnected AI systems.

Daniel Pell, VP — Workday·TechRadar · Workday Research ↗

The report title: The Copy/Paste Economy: Why Task-Oriented AI Is Failing the Enterprise. Workers spend nearly a full working day — 7 hours — per week copying information between AI tools.

Seven hours a week. That's 364 hours a year, per person. For a problem that is fundamentally a routing problem.

What the research told me about what to build

After reading all of this I had a clear picture of what Crossagent needed to be:

  • No re-explaining. The context from one AI has to carry to the others. That's Engram — the memory layer that stores what you talked about and injects it before every message.
  • Stay where you are. People aren't frustrated by having multiple AIs. They're frustrated by having to leave the one they're using. The panel stays in the tab you're already in.
  • Works with existing subscriptions. No one wants to pay again. Native routing uses your own accounts — your ChatGPT Plus, your Claude Pro, your Gemini Advanced. Crossagent costs nothing extra.
  • No new workflow. @mention is already how people think about tagging. It's Discord syntax, it's Twitter syntax. It felt like the right primitive.

The research didn't change what I built. It confirmed that what I was already building was the right thing — and gave me the exact words people use to describe why they need it.

All those public posts are still there. Every one of them is a real person, still doing this manually today.

Written by

Tanmay Sangam

Builder, Crossagent  ·  Northeastern University  ·  Boston

You're probably doing this right now.

Install Crossagent. Stop being the middleware.

Add to Chrome — free →

← PREVIOUS

I built Crossagent in a weekend because I had 7 AI tabs open

NEXT →

How @mention routing works: the technical story