Asking your team "where can we use AI?" usually gets blank faces. Codos, a new startup, flips the question: an AI interviews employees about how their work actually gets done, then finds the repetitive parts worth automating. This post explains Codos honestly and gives you a do-it-yourself interview kit for smaller teams.
What you need
For Codos itself: a company of roughly 300 to 3,000 people. Codos sells a 30-day pilot followed by a monthly partnership, with custom pricing through its sales team. It is not a sign-up-and-go app. Details are at codos.ai.
For the DIY version below: a ChatGPT or Claude account (free works for small runs), 20 minutes per team member, and their consent.
What Codos does, per its own site and listings:
AI interviews. Codos runs "diagnostic interviews" with employees. Its site shows a live listening interface with a microphone, which points to a voice interviewer.
A "company brain". It combines interview answers with sources such as the org chart, meeting notes, Notion, Slack, Gmail, HubSpot and Google Sheets to map people, projects and operations.
Automation shortlist. From that picture it identifies repetitive tasks and the information each needs, and ranks what is worth automating.
Engineers who build it. Codos sends forward-deployed engineers to set up the system and get selected workflows running with your team.
On-premise option. It offers deployment on your own servers for data privacy.
Who built it: co-founder and CEO Dima Khanarin is a former McKinsey consultant. The CTO, Gleb Sidora, previously worked at Meta. The company is based in San Francisco and backed by a16z Speedrun.
Step by step
You can test the core idea yourself. Start with three to five people who do repetitive work.
Tell your team why. Say clearly that the goal is to remove boring work, not to monitor or replace anyone. Get their OK before you start.
Each person runs this interview in a new ChatGPT or Claude chat. It works by typing, or by voice in the mobile app.
You are a friendly workflow interviewer. Your goal is to understand how my
work actually gets done, so we can find boring, repetitive tasks that AI or
simple automation could take off my plate.
Rules:
- Ask one question at a time and wait for my answer.
- Start with my role and a normal working week.
- For each recurring task, ask: how often, how long it takes, which tools
and files I use, where the information comes from, who I hand it to, and
what usually goes wrong.
- Ask for a real recent example of each task.
- Ask which tasks I find most tedious and which need real judgement.
- Don't suggest solutions yet.
- After about 15 questions, or when I type DONE, give me a summary table:
Task | Frequency | Time per week | Tools | Inputs | Output | Pain points.
Do not ask me for passwords, customer personal data or confidential numbers.Collect the summary tables. Each person copies their final table into one shared document. Remove names of customers or anything sensitive.
Run the analysis prompt in a fresh chat with all the tables pasted in:
Below are workflow summaries from [NUMBER] team members at a [TYPE OF
BUSINESS]. Analyse them as an operations consultant.
1. List tasks that appear for more than one person.
2. Score each task 1 to 5 on: hours saved per week, how rule-based it is,
how available the input data is, and risk if the AI makes a mistake.
3. Give me the top 5 automation candidates. For each, suggest the simplest
approach (template, spreadsheet formula, no-code automation, or AI
assistant), what a human must still check, and a 2-week pilot plan.
4. List tasks that should stay human, and why.
Be specific and do not invent numbers that are not in the summaries.
[PASTE TABLES HERE]Pick one task and pilot it for two weeks. Measure the time before and after. Only then move to the next one.
Good to know
Codos is early and enterprise-only. Its customer results, such as a fintech freeing 21% of team capacity, come from the company's own posts and have not been independently verified.
Privacy first. Don't paste customer data, salaries or confidential documents into a consumer AI chat. Check your company's AI policy. For sensitive work, a business plan or an on-premise tool like Codos is the safer route.
Consent matters. Interviews about someone's work can feel like surveillance. Be transparent about what you collect and how it is used.
AI suggestions need a human check. The analysis prompt gives you a starting list, not a final decision.
Sources
Found this useful? Follow @kevinsingh.me on Instagram for a new AI tool every day, and subscribe so you don't miss the next one.