KD Works

Practical help for busy teams

Better tools for the work you already do.

KD Works helps organizations get more out of the tools they already have. We look at how your team uses AI today, help you use it well, safely and together, and build something custom when the tools on the shelf fall short.

Start a conversation See the AI audit

What we do

Ways we can help

Most teams don't need more software. They need the tools they already pay for to work together. Here is where we usually start.

  • AI audit

    We look at how your team uses AI today and how you want to use it, then show where it helps most. How the audit works.

  • Safe use and guardrails

    Clear, simple guidance on what is encouraged, what needs a second look, and what is off limits, so people can use AI with confidence.

  • One team, one approach

    When everyone picks their own tools and habits, work drifts apart. We help your team agree on shared tools, shared prompts and shared rules.

  • AI training for teams

    Hands-on sessions for ChatGPT, Claude, Microsoft Copilot and Gemini. What each one is good at, how to ask well, how to use them safely, and prompts your whole team can share.

  • Workflows and automation

    We map a process your team repeats every week, then build it in the tools you already use, such as intake, approvals, reports and handoffs. For example, tools like Google Drive and Sheets, Microsoft 365 (Outlook, Excel, Teams and SharePoint), Asana and Slack.

  • Email management

    Calmer inboxes. Labels and rules that sort the mail for you, reply templates, daily or weekly priority summaries, and a cleanup routine that sticks.

  • Custom builds

    When the job goes beyond what your tools can do, we build something small and focused, and we match it to the right level of AI. See the ladder.

  • Ongoing support

    Check-ins, fixes and tuning after launch, so what we build keeps working as your team and tools change.

AI audit

Start by seeing how AI is already used

Most teams are already using AI, officially or not. Before changing anything, we find out what is really happening and what you would like to happen. Then we show you how to get the most from it.

  1. How you use AI today

    Which tools people use, for what, and how often, including the tools nobody formally approved.

  2. How you want to use it

    The tasks that eat time, the ideas people have, and what better would look like for your team.

  3. Where it helps most

    We rank the opportunities by how often the task repeats, how much time it takes, the risk, and the value. Some will be a quick fix. Some will be a not-yet.

  4. How to use it safely

    What information should never go into an AI tool, where a person must review the result, and who decides.

  5. How to stay unified

    Shared tools, shared habits and shared rules, so your team works the same way instead of five different ways.

What you get: a plain-English summary of where you stand, a short list of recommended first steps, and a clear view of what is worth doing now and what can wait. If something needs a custom build, we scope it with you.

Custom builds

When the job goes beyond what your tools can do

Sometimes a problem is bigger than the tools you have. When that happens, we build something custom. To keep it simple and safe, we label every project with a level on our Autonomy Ladder, which describes how much freedom the AI is given. We start at the lowest level that solves the problem, and climb only when it needs more. Each level up adds capability, and also cost, delay, unpredictability and testing work.

Workflow · Levels 1 to 6

We decide the steps. The AI does the thinking inside each step. It is predictable, cheaper, and easy to test and explain.

Agent · Level 7

The AI decides the steps, which tools to use, and when it is done. It is the most flexible, and it has the most room for expensive, confident mistakes.

  1. 0

    Product, no AI inside Off the ladder

    Software or a one-time build where the finished product never calls an AI model. AI may help us build it. It is not autonomy, but it is still work we do.

    Example · Landscaping companyA simple online booking page and a spreadsheet turned into a shared crew calendar. AI helped write the code, but the finished tools are plain software.

  2. 1

    Single AI call Most value starts here

    One well-designed prompt, one task, run every time it is needed. Summarize an intake form, draft a reply, classify a document.

    Example · Dental officeEach new-patient intake form is turned into a five-line summary (concerns, history, insurance) for the hygienist before the appointment.

  3. 2

    Prompt chaining

    A fixed sequence of steps. Each step is simple and checkable.

    Example · Accounting firmA vendor invoice arrives. Step 1 pulls out vendor, amount and date. Step 2 checks them against the purchase order. Step 3 drafts the bookkeeping entry for a person to approve.

  4. 3

    Routing

    AI classifies the input and sends it down the right path. Billing goes to one prompt, scheduling to another, and anything unclear goes to a person.

    Example · Veterinary clinicWebsite messages are sorted as appointment request, prescription refill, billing question or emergency. Each type gets its own reply template. Emergencies skip AI and go straight to the front desk.

  5. 4

    Parallelization

    Split the work into independent pieces and run them at once, or run the same check several times and compare for reliability.

    Example · Law officeA vendor contract gets three checks at the same time: one for liability, one for payment terms, one for termination clauses. The results are combined into one summary for the attorney.

  6. 5

    Orchestrator-workers Feels agentic

    A lead AI breaks the job into subtasks it did not know in advance and hands them to worker calls.

    Example · Marketing agency"Build a competitor report for this client." The lead AI decides which five competitors matter and what to check for each, sends each one to a worker, then assembles the report.

  7. 6

    Evaluator-optimizer Feels agentic

    One AI drafts, another critiques against criteria, and the loop repeats until the result is good enough. Use it when "good" is clear but hard to hit on the first try.

    Example · NonprofitAI drafts a grant application. A second AI scores it against the funder's published rubric and lists what is weak. The draft is revised until every section scores well, then a person submits.

  8. 7

    Autonomous agent AI picks the path

    Gets a goal plus tools (email, CRM, web, files) and loops: act, observe, decide the next step.

    Example · Online storeGoal: "resolve this customer ticket." The agent looks up the order, checks tracking, reads the return policy and issues a refund under $50 on its own. Anything bigger goes to a person.

The one question

Can we write the steps down in advance?

Yes: it is a workflow. We build the lowest level that solves it, which covers most business cases.

No: it may be an agent, but only if the path really varies from case to case and a wrong step is cheap or a person approves before anything irreversible.

Start low. We usually start at levels 1 to 3, prove the value in weeks, and move up only when a simpler level hits a ceiling.

More freedom, more guardrails. Limited tool permissions, spending caps, logs, and a person approving anything that touches customers, money or records.

Everyone knows what is being built. Every project gets a level before we start it, so you and we share the same picture of what it does.

How we work

One problem at a time

  1. Listen

    We talk through how the work really happens today, including what's slow, what gets dropped, and what tools you already use.

  2. Find the first win

    Together we pick one problem worth fixing and note where things stand now, so we can tell later whether it helped.

  3. Build or train

    We improve what you have, teach your team to use it well, or build something small if that's what it takes.

  4. Measure and support

    We check the result against where you started, and stay available to keep it running.

What to expect

How we think about the work

Problem first
We start with what's getting in your way. We bring up tools, and AI, only when they help.
Use what you have
Improving the tools you already pay for usually beats adding new ones.
Plain and practical
We aim for things your team can adopt without becoming technical experts.
Measure it
We agree on what better looks like up front, then check whether we got there.

About

A small firm that talks to you like a neighbor

KD Works is a Michigan company started by Dylan Dyjewski and Ian Kumpula. We work the way we'd want someone to work with us: ask good questions, explain things in plain words, and be honest about what will and won't help.

We're a young firm, so you'll be working directly with the two of us.

  • Dylan Dyjewski

    Dylan Dyjewski

    Co-founder

  • Ian Kumpula

    Ian Kumpula

    Co-founder

Contact

Tell us about one recurring problem

A task that eats too much time, a process that keeps stalling, or a tool nobody quite uses. Send us a few lines and we'll reply.

info@kdworksllc.com