Experience over Discourse
Form opinions from actual experience and usage, not online discourse. The closer you are to the systems being built, the more reality diverges from the timeline.
The Thesis
Twitter, LinkedIn, and tech meetups produce discourse - strong narratives that cycle quickly, amplified by small groups of loud voices. Discourse is signal, but it is not ground truth. The opinions worth holding are the ones formed by using the thing, building with it, or talking to people who do.
When discourse and direct experience diverge, trust the experience. The timeline can be wrong about a tool for weeks before the people actually shipping notice.
The Worked Example
Two weeks before this page was written, "MCP is dead" was trending on X, amplified by a small meetup of ~30 people. The timeline had effectively declared a paradigm dead.
After visiting Anthropic and talking to people running these systems in production, the picture was the opposite of the discourse:
- Skills are not a replacement for MCPs - they serve different roles
- Skills are most effective for tight, scoped context and fast iteration
- MCPs provide the backbone: structure, permissions, security
- Anthropic uses skills that route to MCPs - skills as an orchestration layer on top
The "new paradigm" still runs on MCPs, just with cleaner, higher-level control layers built on top. The discourse had compressed a real architectural shift into a false binary. Source: Kevin Liu LinkedIn, 2026-03
Why Discourse Drifts From Reality
| Failure mode | What it produces |
|---|---|
| Hot-take economy | Strong claims travel faster than nuanced ones |
| Small loud cohorts | A 30-person meetup can manufacture a "consensus" |
| Binary framings | "X is dead / X is the future" replaces "X and Y compose" |
| Missing the layer below | People declare the abstraction dead without checking what's underneath |
| Selection bias | The people posting are not necessarily the people shipping |
The Better Loop
- Use the thing yourself. First-person experience is the highest-fidelity signal.
- Talk to the people running it in production. They have the failure modes the timeline doesn't.
- Look at the actual architecture. Discourse argues about labels; systems show how the pieces compose.
- Notice when binary framings appear. "X is dead" almost always means "X and Y are now composing in a way that nobody has named yet."
- Only then form an opinion - and tag it with how you formed it.
Connections
| Connection | Implication |
|---|---|
| Builder Ethos - "Search Before Building" | The cost of checking primary sources is near-zero. The cost of repeating discourse is reputation. |
| Imagination Chasm | Loud framings dominate discourse precisely because they're easier to imagine. Reality is usually compositional and harder to picture. |
| The Brain-Agent Loop | The wiki compounds primary-source signal. Discourse evaporates; compiled experience persists. |
Anti-Patterns
- Restating the timeline as analysis. If you didn't add experience, you added noise.
- Forming opinions from headlines and quote-tweets. The opinion is only as good as the source.
- Treating "everyone is saying X" as evidence. A small group can manufacture "everyone."
- Accepting binary framings. "X is dead" is almost always wrong; "X composes with Y" is almost always right.
- Confusing volume with depth. The loudest opinions are usually the shallowest.
What This Means in Practice
- Before forming an opinion on a tool or paradigm, use it for at least one real task.
- Before repeating a take, identify whether you're sourcing it from a builder or a commentator.
- When the timeline declares something dead, treat it as a prompt to investigate, not as a conclusion.
- Visit the source. Read the actual code, the actual docs, talk to the actual team. This is consistently undervalued and consistently produces edge.
Timeline
- 2026-04-18 | Articulated by Kevin off the back of a visit to Anthropic that contradicted the trending "MCP is dead" narrative. Filed as a standalone philosophy on opinion-formation. Source: User, 2026-04-18; Kevin Liu LinkedIn, 2026-03