TL;DR: Agents now produce a real share of an organisation's knowledge work, and that output has none of the machinery ordinary work has: no history, no diff, no way to move it between tools, no way to show it to someone without a seat. PainHunt's DevTools cluster is asking for a neutral layer that holds it.
The evidence
Developer Tools carries 383 posts scoring 10+/15, average intensity 7.3/10. The source mix is genuinely developer-native — Medium, Mastodon, Bluesky, Discourse — rather than concentrated in review sites.
Four complaints in this cluster describe the same missing layer from different directions. Teams report vendor lock-in preventing them from switching AI models as the frontier moves. They describe the absence of a neutral surface for agent-produced knowledge work, which leaves employees without a licence unable to reach the output. They report that existing solutions lack version control and diffing for AI-generated artifacts. And they describe being unable to import a full agent session — its context and history — from one tool into another.
The requested features name the product almost exactly: an agent-agnostic workspace for HTML and Markdown artifacts, session import with context tracking, collaborative editing that preserves identity, and flat organisation-based pricing rather than per-seat licensing.
That last one matters. It says the buyer wants the whole company to see the output, and the current licensing model prevents it.
Why now
Two years ago an agent's output was a chat transcript you copied somewhere useful. Now it is a document, a page, a plan, a patch — an artifact that other people depend on and that gets revised across sessions.
But the artifact lives inside whichever agent made it. Each tool has its own session format, its own history, and its own seat licence. Switching tools means abandoning context; sharing output means buying a seat for someone who will never type a prompt. Meanwhile the model layer beneath is churning fast enough that nobody wants to be married to one vendor's workspace.
So the artifact — the thing with lasting value — is the piece with the least infrastructure around it. That gap does not usually last.
The wedge
Own the artifact, stay neutral on the agent.
- A workspace for the formats agents actually emit — Markdown and HTML — with real version history and a diff between runs, so a revision is reviewable rather than a mystery.
- Session import: pull an agent's full context and history out of one tool and into the workspace, so the record survives a change of vendor.
- Read access priced at zero. The person who needs to see the output is not the person who paid for the agent, and per-seat pricing is exactly what is blocking adoption.
Land on "diff what your agent changed, and show it to someone without a licence," then expand into review workflows and attribution.
Risks and honest caveats
- Session formats are undocumented and moving. Importing another vendor's context is integration work against a target that changes without notice, and vendors have no reason to help. This is the whole cost of the product.
- The agent vendors will build a workspace. They already have the session data and the distribution. Neutrality is the only durable position, and neutrality is worth less than it sounds until at least two agents matter to the same customer.
- "Version control for AI output" is a crowded pitch and a thin one. Diffing prose that a model rewrote wholesale is not the same problem as diffing code, and pretending otherwise produces a tool nobody opens twice.
- Flat org pricing sounds obvious and prices badly. The buyer wants it because it removes their constraint, not because it maximises yours. Check whether the resulting ACV supports the integration burden above.
How to validate this further
Read the agent-tooling threads in the Pain Point Browser and test whether the neutral-layer framing survives contact with real teams using the Idea Validator. Related: checkpoint and resume for long agent tasks and migrating between LLM vendors behind a gateway.