You Are Ducking Enough: Scope, Craft, and the Small Sparks That Keep Us Going

In The Staff Engineer’s Path, Tanya Reilly points out that a staff engineer's reporting chain determines their scope. If they report to a director, they should clarify whether they’re expected to operate across the entire organization or focus on specific teams or technology areas. If everything becomes the engineer's problem, they’re at risk of having a lack of impact by being spread too thin acoss endless side quests. They also risk a lack of impact if they spend time on work that doesn’t require staff-level expertise. Operating without a clear scope creates difficulty.

Teams can magnify the importance of their own concerns when everyone cares about the same set of things. They can over-focus and forget the rest of the world exists. Loss of empathy makes it difficult for engineers to communicate with non-engineers. Teams can become absorbed in interesting technical details while forgetting users are waiting for the system to be back online. They can stop noticing problems after working around the same broken processes or configurations for months.

Technical vision and strategy documents can be overkill for many situations. A staff engineer should consider what shape the document should take based on their organization’s needs and what it will support. When a lack of direction is slowing progress, they should gather a group to create an abstract high-level vision. Writing technical vision or strategy documents takes significant time.

Staff engineers are expected to tackle problems that seem intractable and make them tractable. A project leadership approach sets up structures that reduces ambiguity and makes it easier to share context across the team. The engineer should proactively build relationships and deliberately work to establish trust. They should write things down, they should be mindful that all decisions involve trade-offs, and they should expect problems to arise and plan accordingly.

A common management challenge is being blocked by another team that’s not delivering needed work. Information gets lost even in organizations with clear communication paths. Life events such as resignations, illness, or sudden leave impact team capacity. The blocking team may have even higher priorities that the staff engineer’s project. The engineer should investigate why the other team isn’t moving forward. The engineer can request something smaller if the team lacks capacity.

Engineering teams often get blocked waiting for key decisions from stakeholders about project direction, architecture, APIs, or data structures. Decision-makers may be waiting for information from their own dependencies before they can decide. To make progress, the engineer should understand and explain, building empathy and remembering they’re on the same side. They should build a mental model of how the other person receives their question and consider potential misunderstandings. When decisions still won’t happen, they can get organizational support by talking to their sponsor about paths forward.

In 1984, by George Orwell, Winston decides to come home and begin the diary because of an incident that had happened that morning at the Ministry. He feels a peculiar uneasiness when a girl he often passes in the corridors is anywhere near him.

Taking leave of Mrs. Parsons, Winston feels as though a red-hot wire has been jabbed into him, spinning round just in time to see her drag her son back into the doorway while the boy pocketed a catapult. What strikes Winston the most is the look of helpless fright on the woman’s greyish face. Back in his flat, a clipped military voice is reading out a description of the armaments of the new Flying Fortress. Eventually, Winston finds himself imprisoned many meters underground. Strapped upright in a chair, a sort of pad grips his head from behind. For a moment, he’s alone.

In Deep Work, Cal Newport describes how Carl Jung kept the key to his retiring room with him at all times. The feeling of repose and renewal that Jung had in his tower was intense from the start. Jung needed to stay sharp to produce a stream of smart articles and books to further support and establish analytical pyschology.

In The Spellshop, Kiela feeds a chicken she acquires, locates her wandering cactus, and picks a couple of tomatoes for breakfast. She calls “Good Morning” to the garden and the forest beyond. She decides to make more jam that day and open the shop properly.

After a storm hits the island and Kiela has taken refuge at Larran’s house, she looks up at the stairs leading to hers. She swallows hard and wishes there were another way up. Glancing back at his house, she notices that he’s already closed the door.

Kiela resolves to have a quiet, unexciting day.Caz recommends making more pine cone remedies. Kiela suggests that they take a break from remedies.

Later, in a cave, Caz tells Kiela she should stay there. The cactus agrees. However, Kiela knows that she has to go back to the shop.

This has been an incredible busy month. The legislative session has ended and most of the really bad bills impacting the LGBTQ+ got stopped or watered down. The two that escaped were the ban on transgender health care for youth, and a fine of $5,000 targeted at drag shows and pride events in rural Utah. I expect the latter to be enjoined by the ACLU of Utah once the governor signs it.

I attended house district officer training, helped pack neighborhood caucus night boxes and met with my house district leadership team to plan for caucus night. I toured the junior high that we’ll be sharing with another district, and got about nine signs out promoting the event. We are expected about 55 people there and we should have about six volunteers and two chairs there to handle the event. Meanwhile, candidates and their campaign managers have been contacting me about speaking opportunities when we Shamrock the Vote.

I had to hit several locations to find a venue to host the Utah Stonewall Dems Oh No They Didn’t legislative review event we had planned to have on March 15. When we found out that Squatters had a minimum fee of $2,500, the board had me look elsewhere.

I contacted Bout Time, and found out the only meeting rooms they had available were in Utah County, a bit too far out of the way. I next followed a lead on Pins and Ales at the West Valley Fair Mall near me, and even though I though the offer was incredibly reasonable, the board voted “No,” because they didn’t want to charge a cover fee.

I really didn’t want to be the one doing this anymore, but eventually I gave in and secured the Second Summit Cider location for March 22. We still plan to celebrate Pi and Queer Day, even though it’s eight days later. We have been getting a good response for both opportunity gift basket donations, and sponsorship donations. We opted to provide three tiers of donations at $100, $200, and $300, with various benefits at each level. Already, this fundraiser appears to be more successful than any other fundraiser we have had in the past, with the Vice Chair Zachary Angel leading the charge, and wihthe assistance of our intern Roman, who had sought us out and graciously volunteered their time. I had the Utah Stonewall Dems bylaws updated on the website, and with the state and county Democratic Parties, and created a spreadsheet for the caucuses to to fill out when they are having their endorsement meetings.

The ACLU board met on Monday, and prior to business, I encouraged people to describe their tactics when dealing with a sense overwhelmment. There were a lot of great creative ways that people shared that I think we all learned from. Tactics included many things from tuning out of social media, going out of country, and just find meaningful alone time. My personal contribution was stepping back and delegating to others to help carry the load. On Thursday, I shared my presentation with the AI Champions group on AI-Human Synergy. I had a one hour presentation to prepare on AI competency and the Human-AI synergetic solution. I opened first with talking about my father's big yellow handled screwdriver, and how even though it was potentially multi-purpose tool, my father, who had learned auto mechanics in high school repairing tractors, and had a twenty year history in the Air Force in reconnaisance photography, always believed in using the right tool for the right job, and had developed quite a tool collection her carried around in his suburban. After his career in the air force, he worked as a contract trailer repair technician, and took great pride in, and received acknowledgement for, the work that he did. My father’s father before him was woodworker by trade and a master craftsman. What distinguished master craftsmen from their peers was an intimate understanding of the material they were working with, the tools they used, and the techniques they applied. Similarly, when working with AI, it is tempting to use it as the big yellow-handled screwdriver as a solution to every possible problem that we may have, but like the big yellow-handled screwdriver, using it nonjudiciously may not produce the quality that we desire or effect we truly desire, and may inadvertently cause damage. My lead, Tony, had already talked about how to be competent with the material. I focused on using AI as a tool, especially LLM-backed agents, like copilot. An agent has three main components: sensors to read the environment, actuators to act on the environment, and a reasoning engine to translate what is read from the sensors to the actions to take on the environment. I talked about how the agent architecture applies to copilot. The sensors are effectively the context, which consists of prompts and anything that is given in context. The actuators are the resulting text from multiple interactions with the LLM and the MCP plugins. The reasoning agent iteslf is an LLM.  LLM is short for Large Language Model. A Large Language Model is essential a multilayer neural network with tens of layers which take the input of each layer, and feed the input with its layer activations to the next layer. The input to the network is the most recent tokens of words in order, with greater weight usually being assigned to the set of most recent words. The internal weights are the entire vocabulary of tokens in the language modified by the location of the token in the context. The output of the neural network is the entire vocabulary with probability of each token being the next one in the sequence.

The final step of the LLM is to pick one of the tokens stochastically (think of it like rolling a thousand-sided die) print it out, shift the input one to the left and feed the new token in. This whole process repeats until the stop token is encountered.

This matters because the neural network does not have any sense of consciousness. It doesn't "understand" what you are asking of it. The LLM was fed the entire corpus of available published English language, plus additional instructions at a moment in time when it was trained. It was trained to generate the next most likely sequence an token at a time, given a specific input.

An LLM is essentially a nondeterministic pattern completer for text. This means that not only can the answer vary, one character difference, like a period at the end of a sentence can alter the generated text. Without a vector store of recently curated data, the solutions provided may be out of date. Because it has been trained on good and bad examples, the solution generated may be wrong or a bad fit.

Context matters a lot. More recent context is given greater weight. Too little context fails to constrain the solution space. Too much context causes the guradrails to lose their solution. Because of its generative stochastic nature, hallucination is a problem. All-in-all, the larger the source file, the time it takes copilot to process, and the more frequently it makes errors, like deleting lines of code.

I offer the following tips. Maintain a procedural context file for long repetitive processes. Refer to the procedural file when asking copilot to apply it on short (500 lines or less) files. In large files, select the lines of code you need it to operate on and use the interactive popup to prompt for the change.

Review every change. Don't commit anything until everything works. This way you have an easy way to determine why something broke, and you can recover easily.

Leverage the tools in your IDE, like running the unit tests, quickly commenting out uncalled mocks, etc. The tools in the IDE are faster and more deterministic.

Finally, have patience with the agent. You will likely find yourself trying to guide the agent. Make sure you ask it to save your guidelines to a context file. After the theroretical presentation, I showed some examples in my workspace and then took questions. The overall response was incredibly positive, with a lot of heart emojis being tossed out, and praise from my lead and my skip lead. Earlier, my skip lead had a one-one with me discussing my upcoming conditional promotion and then said that I needed to fly out to company headquarters sometime in the next couple of month to meet everyone. Meanwhile, I’ve just been assigned to a high-visibility project to replace an integration that is being deprecated by one of our vendors. It was an intense day Friday tackling the project, and we have a follow-up meeting scheduled for Monday morning to try to complete the integration. Yesterday, we had a platform committee meeting for the Salt Lake County Democratic Party. We discussed and recommended two additions to our platform. The first amendment encourages the use responsible and transparent use of artificial intelligence. The second amendment supports honoring due process and other civil liberties in immigration enforcement. Tonight, I attended the kickoff for the American Federation for Society Prevention at the Utah Cultural Celebration Center. I volunteered to table for West Valley Veterans Committee at the event. On the table, in addition to pamphlets on the committee, there were also a few tiny rubber ducks with hats and writing on the front. I picked up a green duck and read “You are ducking enough.” I loved it so much, I brought it home with me and put it on a shelf at eye level where it can give me or whoever encounters it hope when the see the duck and read the words. Tonight I learned that construction industry has one of the highest suicide rates in the country with 56 deaths pre 100,000 male workers and 10.4 deaths per 100,000 female workers. Construction workers are about five times more likely to die by suicide than from on-the-job accidents or injuries. Contributing factors include a “tough guy” or “macho” work culture that stigmatizes vulnerability, high rates of chronic pain and workplace injuries leading to self-medication and opiod dependency, job instability and travel requirements that separate workers from their families and support groups, and the financial stress and high-pressure nature of construction projects.

The lesson threading through all of this—engineering strategy, deep work, activism, storytelling, and even a tiny green rubber duck—is that meaningful impact comes from choosing where to focus and showing up with intention. Whether we’re designing systems, guiding AI, organizing our communities, or simply trying to make the next right decision in a complicated world, the work asks us to balance craft with empathy and persistence with care. And sometimes, when the work feels overwhelming, the spark that keeps us going comes from the smallest places: a kind word, a shared laugh, a moment of recognition, or a tiny reminder sitting on a shelf that we are, in fact, enough. So define your scope, sharpen your tools, and step back into the work that matters most. The systems we build, the communities we strengthen, and the hope we carry forward all grow when we keep showing up—and when we invite others to build that future with us.