ISSUE 001 · THE BEGINNING
How AI gets to work
Welcome to the first issue of Flow & Function (get it: Form and Function?). Each Friday I hope to bring you 3 news items I found interesting from the week and one real workflow you can both test and put into practice. This week is a little different. Instead of sharing a workflow, I want to introduce some of the terminology we’ll be using.
News: 3 things worth knowing
Who gets to write the rules for AI?
OpenAI has launched AI Futures, a new internal team focused on how society can keep increasingly powerful AI from concentrating too much control in the hands of governments or a small number of companies. Its first essay raises difficult questions about personal freedom, privacy, accountability, and keeping people, and not machines, in charge of important institutions. It’s worth keeping up with the group and their writing, particularly because Axios previously reported that the team’s leader, Dean Ball, would work on both public policy and OpenAI’s internal governance.
Consider: OpenAI is studying how to prevent concentrated power while being one of the companies accumulating it. An internal team could help the company make better decisions, but it could also give OpenAI more influence over how the public talks about regulating AI. So what comes next? Does the AI Futures team produce concrete ideas that lawmakers, researchers, and the public can challenge and use, or will it remain a thoughtful conversation inside the very company it may need to challenge?
AI jobs might include a hard hat
Meta says the first participants have graduated from America’s Workforce Academy, its short, no-cost training program for construction and fiber jobs connected to data centers. The company covers training, travel, lodging, and a stipend, and says graduates are guaranteed work with a Meta partner or at one of its construction sites. Independent reporting from Axios also notes that graduates receive an industry-recognized credential from the NCCER. The program is a useful reminder that some jobs created by AI may involve cables, cooling systems, and construction. This is a positive reskilling initiative, and I’m even thinking, “Gee, maybe I should learn to be a fiber technician.”
Consider: The complication is that these data center opportunities depend on what actually gets built. Communities across the country are pushing back on data centers over electricity prices, water use, tax breaks, noise, and local control, while some states are considering tighter limits or requiring developers to cover more of their infrastructure costs. New York recently became the first state to issue a moratorium on data center construction. Recent Associated Press reporting shows how quickly the political landscape is changing.
The bigger reskilling question here is whether these credentials can lead to lasting, portable careers in spite of the pushback on the structures throughout the country.
ChatGPT for Teens adds guardrails
ChatGPT for Teens gives users ages 13–17 stronger default protections. OpenAI says the teen experience places tighter limits on conversations involving self-harm, violence, eating disorders, sexual content, and emotional dependency. It also includes break reminders, homework and study features, and reminders not to share sensitive images. When a parent and teen link their accounts, parents can set quiet hours, manage selected settings, and receive limited alerts about certain high-risk situations without routinely seeing the teen’s conversations.
Consider: Those protections arrive as states are beginning to turn some chatbot safeguards into legal requirements. The Transparency Coalition’s guide connects the product launch to laws covering areas such as disclosures and reminders that the chatbot is not human, appropriate responses to signs of self-harm, and protections against sexual content for minors. The laws vary by state right now, and not all of the included features are required in every state.
Will age detection, guardrails, and parental controls/alerts work reliably in ordinary use? It’s a meaningful start, but families still need evidence that these guardrails will hold up in practice.
Workflow of the Week: What are we talking about?
This is where we’ll normally have a workflow or workflow related item that you can create and/or test. But before we build anything, let’s get our vocabulary straight.
You already have workflows: applying for a job, approving an invoice, planning a meeting, publishing a newsletter. Each one begins with something that needs attention and moves through a series of steps toward a result. Some people draw a distinction between a process and a workflow, but for our purposes, they’re interchangeable. Both describe how work moves from a starting point to a finished outcome. Workflow simply helps us focus on the actual sequence: what happens, in what order, who or what handles each step, and how we know the work is done.
Before improving a workflow, it helps to dissect what really happens as opposed to outcome, which is what the instructions say should happen. What starts the work? What information comes in? Which steps repeat? Where does someone make a decision? Where does the work wait for another person, move between apps, or get checked? What is the final result? Breaking the work into those smaller parts is called decomposition. It turns a vague activity such as “track my job search” into visible pieces that can be examined and improved.
Find the friction before choosing the tool
Once the steps are visible, look for friction: anything that makes the work slower, harder, less reliable, or easier to abandon. Friction might be copying the same information into two places, hunting for a document, waiting on an approval, forgetting a follow-up, reformatting information, or checking an AI-generated answer. Friction is annoying, but not all friction is bad. A deliberate review before sending a contract or submitting an application is a safeguard, not waste. The goal is to remove needless effort without removing useful judgment.
Keep in mind that a workflow does not automatically need AI. Sometimes the best improvement is a clearer order, fewer handoffs, one shared document, a form instead of an email, or a calendar reminder. This is why decomposition is so helpful. Identify the friction and add AI only when it is useful for work (e.g., summarizing, extracting information, classifying, comparing, or drafting), and only when mistakes can be caught before they cause harm.
Key terms to know about workflows
A workflow is an entire repeatable process, from beginning to end.
Automations run one or more predictable parts of it: when this happens, do that. If X, then Y.
An AI agent works toward a goal across several steps and makes decisions within the boundaries it has been given. Depending on the task, it may still require human approval.
A connector, integration, or plugin within an LLM gives a tool permission to work with another service. For example, within ChatGPT or Claude, you can connect items such Google Drive and apps, Microsoft apps, Slack, Notion, Dropbox, CRMs, as well as tools like Zapier. The names vary by product, but the basic job is the same: let information or actions pass safely between tools without constant copying and pasting.
A few more terms worth knowing
Input: What enters the workflow, such as a form, email, file, request, or idea.
Trigger: The event that starts a step, such as receiving an email or reaching a due date.
Decision point: A place where the path changes and based on the content: approve or revise, pursue or pass, routine or exception. Some decisions require human judgment.
Handoff: When responsibility or information moves to another person or tool. For instance, when ChatGPT hands off information to Zapier.
Source of truth: The one place everyone agrees contains the current, reliable version.
Output: The useful result the workflow is meant to produce.
Human review: The point where a person checks accuracy, judgment, tone, risk, or permission before work continues.
Exception: A case that does not fit the normal path and needs special handling.
You don’t need to memorize the vocabulary. The practical idea is simple: first understand the work, then find the friction, and choose the tools last.
Next week, we’ll put that into practice with a job-search tracking workflow that starts with a saving a job posting and ends with an organized spreadsheet and clear next actions.
One question to take with you: Where could a small, repeatable AI workflow remove friction from your week?
Until next time,
Michelle
Flow & Function: AI in Working Order
AI Disclosure:
This newsletter uses AI for the following: design, research, initial summaries, and finding multiple sources. The final copy, writing, perspectives, and workflow design are created by Michelle.