Spencer Hardwick

Spencer Hardwick

AI product manager who builds

I turn questions about agents and LLMs into working products. I own the work from product direction through architecture, implementation, and evaluation.

Black-and-white portrait of Spencer Hardwick.
Explore my work

Selected work

Product direction through implementation.

Customer workLive in production

Customer operations agent

Give the model room to interpret. Make actions reviewable.

A production agent that helps a customer team prepare operational work through conversation, review a concrete proposal, and carry out approved actions. I own the product and technical work end to end, including its continued development around real use.

Product and technical owner

Read the case study: Customer operations agent
The person reviews a concrete proposal before the application acts.

  1. Prepare

    Conversation helps the operator express a request and prepare a concrete proposal. The proposal gives the person something specific to inspect.

  2. Review

    A person reviews the proposed work before it proceeds. Understanding a request and authorizing an action are separate responsibilities.

  3. Act

    Application code carries out approved work and returns a report. The model's interpretation alone does not authorize the consequential action.

Independent projectShelved

Calx

When a repeated correction needs a mechanism.

Calx grew from a question in my own agent-assisted builds: what changes after I correct an agent? I developed a platform for capturing recurring corrections, proposing rules, and enforcing approved behavior in the runtime. The work spans practitioner research, an early CLI, and a later platform prototype.

Founder; product and technical owner

Read the case study: Calx
Recurrence suggests a rule. The user decides whether to promote it.

  1. Capture

    Record a correction with enough context to describe what happened and the behavior that needs to change.

  2. Recognize recurrence

    Compare corrections to identify a repeated issue. Recurrence is a reason to consider a structural change, rather than proof that a rule is ready.

  3. Suggest a rule

    Surface the repeated issue as a candidate for enforcement. The suggestion gives the user a decision to make.

  4. User promotes

    The user chooses whether to promote the correction. Crossing a recurrence threshold alone does not start the compilation process.

  5. Validate

    The compilation path generates and validates an enforcement artifact before distributing an approved rule to the runtime.

  6. Enforce in the runtime

    Middleware applies the rule as the agent works. The application can govern tool availability and actions outside the model's conversational context.

Independent projectShelved

Versed

Make the translation visible.

Versed explored the gap between how candidates describe their experience and how employers describe a role. I moved the product from a ranked job feed toward visible, grounded translations that users could inspect. The product is shelved; I still reuse parts of it in my own job search.

Founder; product and technical owner

Read the case study: Versed
An illustrative mapping from existing experience to employer requirements.

  1. Candidate evidence

    Start with the candidate's existing experience. A translation should preserve what the person actually did and make its relevance easier to understand.

  2. Employer requirements

    Identify the language and competencies in a job description. The catalog provides grounding for relationships between candidate and employer vocabulary.

  3. Visible translation

    Present the mapping as a structured artifact, with source evidence and gaps available for review. The explanation becomes part of the interface.

About me

I work across product and implementation, using coding agents throughout the build. My projects explore how people understand an agent’s work, where software should control its behavior, and how to verify the complete experience.

I’m also a multidisciplinary artist. That practice sits alongside my work in technology.