At Bedrock, my daily use of AI goes far beyond code
When AI comes up in Tech roles, the conversation quickly turns to code. In my day-to-day work at Bedrock Streaming, much of the value is elsewhere: keeping track of context, preparing decisions, and making execution easier to understand.
The daily problem is first a context problem
In a sizeable Product & Engineering organization, information is distributed across Slack, email, calendars, meetings, Jira, and working documents.
The risk is not only missing one piece of information. It is losing the links between them: a decision made in a meeting, a signal appearing in Slack, a commitment mentioned by email, and a task that eventually needs to be tracked in Jira.
AI helps me reduce the cost of rebuilding that context.
I deliberately stay at the method level in this article: Bedrock’s internal content obviously remains internal.
Slack, email, and calendar: finding signal in the noise
I use AI to summarize discussion threads, connect information spread across channels, and prepare the topics that actually require my attention.
For the calendar, it helps me prepare a day or a week: which meetings require preparation, which topics are connected, and which decisions are still open?
For email and Slack, the goal is not to summarize everything. It is to identify changes, decisions, risks, and requests for action that deserve to surface.
A poor summary creates another layer of noise. A good synthesis system reduces cognitive load.
A meeting is only useful if it produces a clear next step
Meeting notes are another very practical use case.
AI can turn notes or a transcript into an actionable structure: decisions made, open points, owners, deadlines, and next actions.
I do not treat that output as automatic truth. I review it, correct ambiguity, and verify important commitments.
The value comes afterwards: everyone starts from a more consistent level of information and it becomes easier, a few days later, to check whether a decision really turned into action.
Artifacts make abstract problems visible
I also use Claude to create artifacts quickly when I need to make a problem tangible.
One example is Jira data quality. When a process depends on fields, statuses, or links between issues, it is hard to discuss its implementation based only on impressions.
An artifact can turn that data into an understandable view: completion rates, inconsistencies, items without owners, missing stages, or change over time.
The goal is not to build one more dashboard. It is to create a discussion support quickly enough to understand whether the process we defined actually exists in the data.
Measuring a process is better than assuming adoption
A Product & Engineering transformation does not succeed because a new process has been presented.
You need to observe adoption and understand where it breaks down.
AI-assisted artifacts and analysis help shorten the gap between a hypothesis and verification. We can test a metric quickly, see whether the data is usable, identify gaps, and then decide whether that indicator deserves to be industrialized.
That also helps avoid building permanent reporting too early around a poorly defined metric.
Guardrails remain human and organizational
These use cases do not mean sending every piece of company information indiscriminately to any tool.
Tool selection, confidentiality rules, access rights, data sensitivity, and output validation remain normal organizational responsibilities.
Likewise, I do not delegate a decision because a model produced a convincing summary.
AI gets me faster to the point where my judgment is useful. It does not replace that judgment.
What this changes in my role
Daily AI use makes me more hands-on, but not only around code.
I stay closer to the organization’s real flows: information, decisions, meetings, data, tasks, and adoption measurement.
For a CTO / CTPO or transformation leader, that may be one of the most interesting changes: AI lowers the cost of accessing detail without forcing you to abandon the wider view.
The challenge is to build your own guardrails so that this additional proximity improves decision quality rather than simply increasing the volume of information produced.