TL;DR — Q2 shipped three client releases, two in-house product updates and laid the groundwork for our content pipeline. We missed our cmi5-support estimate and learned to validate before quoting. Q3 priorities are clearer because of it.
Three months ago we set a Q2 plan. Two months ago we hit a thing that was harder than expected. One month ago we shipped most of it. This is the honest version.
What shipped
- Blink AI — switched the underlying frontier model with no user-visible drift (how), added per-user memory isolation tests to CI
- Nexia Academy — three new tenants live, the cmi5 launch flow fixed for one customer LMS (after a hard week)
- Wallnetic — multi-display refresh-rate fix shipped, App Store approval secured for the new entitlements
- Client work — three production releases, all on schedule, two with on-call coverage we operate
What we got wrong
- cmi5 LMS support estimation. We took a vendor's docs at face value. Operational support did not match. Three-week support thread, frustrated customer, eventual workaround. We now validate before quoting.
- Q2 blog publishing cadence. Started strong, dropped off in mid-Q2 when client deliveries spiked. The Q3 calendar treats blog as a 4-hour-per-week budget item, not a “when there is time” task.
- Internal tools investment timing. Should have built the multi-tenant admin two months earlier. Cost ~10 hours of operator time we could have avoided.
What we paused
- A second native macOS product — scoped, prototyped, then deferred to focus on Wallnetic stability
- Public API for Blink AI — design is done, build deferred to Q3 to avoid context-switching mid-quarter
What carries into Q3
- Blink AI public API ship
- Multi-region Postgres pilot on Hetzner
- Nexia Academy reporting layer expansion
- A documented cmi5 pre-flight checklist available to customers (now public)
Frequently asked questions
Why publish a quarterly review?
Public retrospectives keep us honest with ourselves and give prospective clients a real signal — they can see how we work, what we ship, and what we admit when something is harder than expected.
What was the biggest miss this quarter?
Underestimating cmi5 LMS support fragmentation. We promised a feature based on a vendor's docs and learned the hard way that documented support is not operational support. We now validate every cmi5 LMS end-to-end before quoting it.
Working on something similar?
T-Square architects, builds and operates production systems for learning, AI and custom software products. Talk to a senior engineer for a second opinion.