I consume a lot of tech blogs about engineering, which discuss how developers optimise their workflows in the age of LLMs. There seems to be a growing trend in improving productivity and output by engaging with coding tools such as Cursor, parallelising workloads, and chaining together agents. I'm still of the opinion that being agentic as an individual is the primary separator between those who can single-handedly impact a piece of software, and those who are merely contributors.
So, what is being agentic? It does not mean embracing AI in every part of your life. For me, it is the determination and willpower to drive change consistently in areas that others either struggle with or are not willing to engage with. These areas are usually difficult because of bureaucracy, technical complexity, legacy decisions, or group-think. They have no simple solutions and they demand research and time.
Yet, it seems like there has never been a better time to utilise an agentic mindset with the AI tooling available. Want to understand an obscure domain or architecture? Load up Cursor, set a model to thinking mode, and ask it to dissect the structure and flows that an application offers.
Need to learn a new subject? An approach I've found beneficial is to ask ChatGPT 5.1 to link to a number of sources related to the topic. Once I have read these, I then ask it to design 10 questions which test my core understanding. Any fundamental mistakes - repeat the process. I've begun to do this while I redecorate my house using voice mode (I think the current limit for the Pro plan is 1 hour per day).
Want to make a PR description easy to grok? Ask your model of choice to generate a MermaidJS diagram, highlighting the main changes (this is supported in Markdown). Checking out a branch and asking an agent to use the diff to explain the changes made, and highlight any high-risk edits removes computational load for other areas.
LLMs have also become my sanity check. Given any problem, I first formulate a core strategy. Be it a coding task, this may take the form of an instruction list of the required changes to the codebase or some pseudo-code. If it's a design document, an overview of the approach I intend to take. A prompt I like to use is:
I work for a Work OS company focused on project planning and execution. Given the overview of X below, analyse the approach and suggest improvements. Ignore minor wording edits; focus on the core design and risks. If it's code, point me to similar open-source patterns with links. If it's a design, call out missing steps (such as approaching stakeholders) and early steps that would de-risk the work.
We have long ago passed the point where AI tooling is a gimmick. The real gap is between people who talk about productivity and people who own problems, use the tools ruthlessly, and ship. Talk is cheap.