Pair a useful AI example with a way to check it
Jasmine Solomon starts with staff readiness. A new AI feature does not automatically give people the judgment to use it well. She describes teaching prompt design, checking outputs and gathering colleagues who already have practical examples to share.
Leah Stiff adds a learning use case: ask for a different explanation, an illustration or a step-by-step account of what code does. The value is the chance to keep questioning a confusing idea. A fluent answer still needs a check against reliable documentation or a small known example.
Choose one low-risk task in an approved tool, using fictional or appropriately cleared material. Write down the expected result and the errors a reviewer should catch. Have the learner explain the answer back. Reusable prompts can reduce friction, but they should carry the checking habit with them.
In the recording · 03:12 ↗Give a promising feature one concrete job
Dustin Gerstenfield describes a small team interested in AI-assisted portal building after a rapid system transition. The appeal is practical: turn an idea into something usable when staff time is scarce. His comments reflect the features being discussed at Summit 2025, rather than a completed portal demonstration.
For a similar team, begin with one user task and an example of successful completion. Identify the fields, access rules and result the first version needs. Try that narrow workflow with a colleague before adding more screens or design flourishes.
Check how the team will handle an incorrect result, a missing input and a future change. Keep time for testing and maintenance in the plan. Generating a first version can help a team get started; someone still needs to understand the system it becomes.
In the recording · 11:26 ↗Borrow a better data structure, not just a visual idea
Jonathan Thornton sees a new possibility in colleagues using datasets to build event agendas, including events that do not require registration. His own scheduling work gives him a concrete problem to reconsider. The useful discovery is how the information is organized behind the page.
Before copying an agenda design, list what a record represents: an event, a session, a room, a time slot or an attendee registration. Decide which relationships the page actually needs. An informational schedule and a registration workflow may share a visual layout while requiring different behavior.
Sketch a small example with overlapping sessions and a room change. Ask who edits the information, how the update reaches visitors and what the page should show when a session is unavailable. Test the structure before investing in the final presentation.
In the recording · 13:29 ↗Make identifier relationships explicit
Robbie Cozad describes bringing data together across different stages of a student’s journey. The systems involved do not always share one consistent identifier. His team uses a dataset of associated IDs to connect those records.
Treat an identifier as a value with a source, not just a string. The same number can mean different people in two systems. Define what each row represents and which relationships are expected to be one-to-one or one-to-many before using a mapping in an import.
Test a known match, an unmatched value and a conflicting mapping. Route ambiguity to review instead of silently attaching the data to the first result. Record who owns the mapping and how a correction affects downstream records. A useful crosswalk makes uncertainty visible as well as connecting the straightforward cases.
In the recording · 15:33 ↗Give governance a decision route and a conversation route
Megan Miller asks who is running the system and how competing priorities will be resolved. Without a clear answer, administrators can work against one another, releases can surprise users and workarounds can grow outside the shared process.
Name who can decide, who needs to advise and how the outcome reaches the people doing the work. Keep a place for requests and problems to move upward as well. A roadmap is useful only if users can understand what is changing and where to raise a concern.
Megan suggests approachable channels such as newsletters, information portals and recurring question sessions. Use repeated questions as evidence: does the team need a refresher, a clearer instruction or a different process? Give someone responsibility for taking that pattern back to the decision-makers.
In the recording · 19:07 ↗Keep improvement work visible between cycles
Peter McKay encourages practitioners to revisit their assumptions as Slate and its community practices evolve. He also recommends keeping a cycle-preparation list throughout the year, while the reason for each change is still clear.
Record the problem, its impact, dependencies and a useful example when it appears. Decide which fixes are urgent and which need a planned review window. Before adopting a community approach, check that it fits your data model and current product behavior.
Keep integration requirements visible without letting one destination’s file layout dictate every internal relationship. Validate the transformation with a sample export and the receiving system. Sustainable work combines a model people can understand with evidence that the complete process still works.
In the recording · 26:36 ↗