Studio Update: What Coldabry Shipped in Q2

A quick roundup of what we built this quarter. We keep these honest: real projects, real timelines, and what we learned. The theme this quarter was speed with discipline. Several builds went from kickoff to production in three to four weeks without skipping the boring operational parts that keep them alive.

For companies

A few highlights from the company side of the studio.

A support agent that deflected most of the inbox

A growing SaaS team had two people stuck answering the same dozen questions all day. We built a grounded support AI agent in front of their helpdesk and cut repetitive ticket volume by roughly 80%, with 24/7 first response. We wrote the whole thing up in the support agent case study, including the parts that were harder than expected, like cleaning up stale docs before the agent could ground on them.

Invoice reconciliation, automated

A finance team was matching payments to orders by hand across two systems that disagreed about everything. We replaced it with an agentic workflow that matches the easy 90%, flags the messy 10% into a review queue, and leaves a full audit trail. Their close got faster and the manual grind disappeared. This is the kind of internal tool we love to retire, and we wrote about the pattern in agentic workflows for internal tools.

A custom ops app in three weeks

A logistics company needed a custom scheduling tool that no off-the-shelf product fit. Classic build-not-buy situation. We shipped a working app in three weeks using our agent-assisted process, deployed with monitoring and rollback from day one. The details of how that timeline is even possible are in ship custom apps in weeks.

For communities

We build for communities too, not just companies, and this quarter had a good run of it.

A member onboarding agent

A large online community was losing new members in the gap between signing up and getting set up. We built an onboarding agent that walks new members through setup, answers their first questions from the community's own knowledge base, and escalates anything human to a moderator. Time-to-first-value for new members dropped sharply, and the mod team stopped answering the same three questions a hundred times a week.

A community knowledge tool

Another community had years of valuable discussion buried in an un-searchable archive. We built a grounded search and answer tool over their history so members can actually find what was already solved instead of re-asking. The best part is it gets more useful as the archive grows, instead of less.

What we learned this quarter

A few things that held up across every project.

Lesson	Why it mattered
Sharp scope beats big scope	Every fast project had a one-page spec. Every slow one did not.
Grounding is the real work	Agents are only as good as the docs and data they sit on.
Escalation is a feature	The projects that worked all had a clean hand-off to a human.
Deploy is part of the build	Monitoring and rollback on day one, not as an afterthought.

The through-line: the AI part is rarely the hard part anymore. The hard part is choosing the right thing to build, grounding it in real data, and shipping it with the discipline to keep it running. That is where the studio spends its judgment.

What is next

Heading into next quarter we are doing more of what worked: tightly scoped builds, agentic workflows that replace manual ops, and grounded agents for both companies and communities. We are also writing more of these posts, since explaining the work clearly tends to attract the people we most want to build with.

If you have a process that should be an app, or a manual job that should be an agent, talk to us. We will tell you in a week whether it is a quick build or a real project, and we will be straight with you either way.