What AI writing tool actually works for technical documentation?

Having evaluated numerous AI writing assistants for the purpose of generating and refining technical documentation—from API specs to internal runbooks—I've found the landscape to be surprisingly sparse in tools that handle technical depth accurately. Most general-purpose writing AIs excel at marketing copy or simple summarization but falter when confronted with precise terminology, code snippets, and logical sequencing inherent to our field.

The core challenges for technical documentation are distinct:
* **Conceptual Integrity:** The tool must maintain strict consistency with defined system components and their relationships. Hallucinating a non-existent parameter or misstating a causal dependency is catastrophic.
* **Code-Awareness:** It must understand the syntax and, ideally, the semantics of code blocks within the narrative, offering suggestions that improve clarity without breaking functionality.
* **Tone Preservation:** Technical documentation requires a consistent, neutral, and imperative tone. Tools that inject subjective language or vary sentence structure excessively degrade readability.
* **Structured Input Handling:** The ability to ingest a bullet list of points or a poorly drafted internal comment and output a coherent, structured section (e.g., a troubleshooting step) is critical.

In my workflow, I've tested these tools by providing them with a deliberately fragmented draft of a database connection troubleshooting guide. The results highlight the divergence:

```markdown
// Input Fragment:
"Client gets timeout. Check network. Verify credentials. Maybe connection pool is full? Look at `SHOW PROCESSLIST;`"

// Desired Output Structure:
"**Symptom:** Client connection timeout.
**Diagnostic Steps:**
1. Verify network connectivity between client and database server (e.g., using `ping` and `telnet `).
2. Confirm the supplied credentials are valid and have the necessary privileges.
3. Investigate connection pool exhaustion. On the database server, execute:
```sql
SHOW PROCESSLIST;
```
A high number of connections in the `Sleep` state may indicate pool misconfiguration."
```

General-purpose rephrasers often fail at step 3, either by mishandling the SQL command format, failing to explain its relevance, or inventing incorrect interpretations of the `SHOW PROCESSLIST` output. The tool that can correctly expand, structure, and contextualize this input—while treating the code block as a sacrosanct entity—is the one that provides genuine value.

My current assessment is that no tool is a complete solution, but some are viable as advanced editing partners. I'm interested in experiences from others who have integrated an AI writing assistant into their technical doc pipelines. Specifically:
* Have you found a tool that reliably respects and correctly formats code blocks or configuration snippets?
* Which assistants allow for the creation of custom style guides or terminology banks to enforce conceptual integrity?
* Is the primary value in drafting from scratch, or in rewriting and clarifying existing, rough text?