~/keystrawai / insights · ~/insights/

Field notes from local-business engagements.

What we ship, what it cost, what worked, what didn't. Specific shops, specific numbers, specific tradeoffs. The opposite of a buzzword blog: short, technical when it needs to be, and only published after a build has closed and the owner has signed off.

// empty-state

Nothing here yet.

The first field note lands when the first engagement closes. We'd rather publish nothing than publish anything generic, so this directory will stay quiet until there is something specific worth reading.

If you want a note when one is published, email hello@keystrawai.com. We add you to a small list and send only when something ships. No newsletter. No drip. No "best of 2026" recap.

subscribe-by-email → start the intake →

zsh ~/insights · 2026-05-02

$ ls -la ~/insights
total 0
drwxr-xr-x  2 ks  staff   64 may  2 09:08 .
drwxr-xr-x  9 ks  staff  288 may  2 09:08 ..

$ cat policy.md
# publish only when:
  - engagement is closed (day 30+)
  - owner has reviewed + signed off
  - real numbers, not estimates
  - we'd send it to a friend asking the same q

$

// backlog

What we expect to write about.

Topics already on the list. Drafted from real engagements in flight; published once cutover reports are signed off. Order is intent, not commitment.

soon · Q3 2026 After-hours intake at an HVAC shop: what the pilot week of texts actually booked. customer-ai
soon · Q3 2026 The cheapest reporting build we've shipped, and what made it usable on day one. reporting
soon · Q3 2026 When NOT to add an AI phone agent, in three live examples. customer-ai
later · Q4 2026 Dispatch logic for an HVAC outfit: rules versus a model versus the dispatcher. ops-scheduling
later · Q4 2026 Tone cloning for proposal drafts: what's worth tuning, what isn't. documents-content
later · 2027 A year of run costs: where the bills actually land for a shop with three integrations live. economics

// editorial

Why the policy is "publish nothing until it ships."

AI commentary is cheap to generate and exhausting to read. The world doesn't need another "10 ways AI will change [industry]" post, especially from a practice that hasn't shipped at that industry yet.

Field notes here will be the opposite: a single shop, the integration we built, the number it moved (or didn't), what we'd do differently. If we don't have one of those to publish, we publish nothing. A quiet log is more credible than a loud blog.

When a note does appear, it will include:

the shop's name (with permission) or a clear anonymization
the integration shipped and the stack it runs on
the pilot + first-week numeric report, unedited
what we'd do differently if we shipped it again
a contact path to the owner if you want to ask them yourself

// next

While the directory is empty, the work is happening.

If you want to be the engagement we write up first, send the intake. The first three intakes get an unusual amount of attention.

start-intake → read the approach →