Kickoff · Est. 3 min · On track

IT Project Manager, New York

  • 6+ Yrs IT Delivery
  • 20+ Projects Delivered
  • Up to 15 Team

PMP, PSM I. 10 years running delivery: almost seven for web, design and AI-native product teams, and before that for live TV shows.

I work with AI agents every day, and also without them.

Kickoff · Project charter

Anton Rybalchenko

Plan · Scope statement

What I own

Own delivery end to end.

From RFP and SOW to release, and the support that keeps clients for years.

Run financial governance.

Monthly budget reports and invoices, so an overrun shows up before it lands.

Build the PMO system.

Processes and AI workflows in an ISO 27001 environment, with a human at every gate.

Bridge business and engineering.

US founders and VPs on one side, engineers in Europe on the other. I keep them in sync.

  • 6+ Yrs IT Delivery
  • Up to 10 Parallel Activities
  • Up to 15 Team
  • 20+ Projects Delivered
Ten lines of work converge on one point.10 of 10 landed.

Watch them still land.

Core skills · WBS

1Delivery

  1. 1.1Stakeholder Management, Initiate to Close
  2. 1.2Agile Delivery, Plan to Monitor
  3. 1.3PMO Governance, Plan to Close
  4. 1.4Risk and Budget Control, Plan to Monitor
  5. 1.5Account Management, Execute to Close
  6. 1.6RFPs and Proposals, Initiate
  7. 1.7Vendor Management, Plan to Execute

Tools

  • Jira
  • Jira Service Management
  • Confluence
  • MS Project
  • Azure DevOps
  • Excel
  • Figma
  • Slack
  • Teams
  • GitHub
  • GitLab
  • Google Workspace
  • Claude Code
  • ChatGPT and Codex
  • Gemini

Plan · Critical path

Role history. What shipped.

Almost seven years in IT delivery at Sigma Software, a company of 2,000+ people. Before IT, I co-founded an event production agency.

Project Manager

Sigma Software · New York · Jul 2022 - Sep 2026

“I would work with Anton again.”

Dmytro Kapustianskyi, VP, Experience Design
  • Ran up to 10 parallel activities for US and European clients: builds, support, RFPs and sales.
  • Owned pre-sales, PERT estimates, SOWs and month-end invoicing.
  • Took over a senior manager's scope and led 15 developers across two teams.

Junior Project Manager

Sigma Software · Kyiv · Nov 2019 - Jul 2022

  • Ran every Scrum ceremony for a cross-functional team of 12 and tracked velocity and earned value.
  • Swedes for Ukraine: took over a platform where Swedish families offered homes to Ukrainian refugees. 1,500 found housing. Sigma credits me in the published case.

CR-001 Change request: from live TV to IT. Status: approved.

Co-Founder and Project Manager

Way Creation · Kyiv · Jan 2016 - Nov 2019

  • Executive producer of Miss Ukraine 2017 and 2018, a live national TV final with up to 10 star performers.
  • In 2017 a giant kinetic hand carried the stars across the stage. One live date that could not move.

Deliver · Benefits register

Business impact.

Every number has its math one click away.

  • to a live healthcare marketplace, against the estimate for a custom build · Team result

    In production in about 9 weeks, against a 5-month estimate for a custom build. Then a release every two weeks. Team result, published case, 2026. My part: nothing new entered an open release.

  • rework after design handoff, by logged hours

    Rework cut about 20%, by logged hours. Designers rated our process 7.3 out of 10, developers 3.5. Then developers started reviewing designs before clients saw them. My result, 2025.

  • managers to run a design request queue for 40+ stakeholders, unchanged four years later

    Management overhead cut 50%. One manager now runs a design queue that needed two. Rebuilt in Jira Service Management, unchanged four years later. My result, 2022.

  • the budget cap on an AdTech website rebuild, with full scope

    An AdTech website, redesigned and rebuilt 20% under its budget cap. Full scope. When the first concepts kept missing, I added a second designer. The buffers in the estimate went unused. My result, 2025.

  • a client who left after a budget overrun came back to me and stayed three years, with ~$200K of new work in a single year

    A client had left after a budget overrun. When I took over the account, they came back and stayed three years. In one of those years they bought about $200K of additional design and development work. My result.

  • “...a pace that would normally require a much larger organization.”

    Lucas Langen, Co-Founder & CEO, DripDrip Pharma

Deliver · Decision log

Four projects. Four hard calls.

From my work at Sigma Software. Client names are left out.

Shipping a healthcare marketplace in nine weeks with an AI-native team

Healthcare · B2B marketplace · Claude Code

A European healthcare company needed a B2B marketplace where licensed doctors order medical products. Everyone on the project worked with Claude Code agents, on our side and on the client's.

The hard part. The team built fast, but new ideas arrived faster than anyone could review them.

The call. Nothing new enters an open release.

If it truly cannot wait, something of the same size comes out.

Result. In production in about 9 weeks, against a 5-month estimate. About 500 doctors registered in the first weekend. Team result.

What I'd do differently. Measure where work waits from week one.

Try it. Drag the new request into Release 1.

Getting designers and developers to stop blaming each other

Design and build · Process · Jira

Designers felt their intent got lost in the build. Developers felt they came in too late. Each side was sure the problem was the other.

The call. Conversation over documents: developers now review the design before the client sees it.

Result. Rework after handoff fell by about a fifth, by logged hours. My result, 2025.

What I'd do differently. Agree the metric with the leads before the change.

Rescuing a 3D website that failed its speed test at acceptance

Web · 3D and animation · Next.js

At acceptance, the client asked for the green zone in Google PageSpeed, a target that did not exist when we started.

The call. I did not argue. Performance was never agreed before the build, and that gap was ours.

Result. 90+ in PageSpeed in two weeks. Now performance, security and accessibility targets are agreed before any build starts.

Getting six country offices to speak with one voice

Multi-site program · Six countries · SEO vendor

A client liked our redesign and gave us more: a site for each of its offices in six countries, plus SEO with our partner agency. One client turned into six.

The call. One voice: each office names one person who decides, and the head office is always in copy.

Result. We made the Monday trade show. One project grew into six, and the program still runs.

The week of the trade show I lost my own voice, so I ran the whole push in writing.

What I'd do differently. Write the one-voice rule into the first contract.

Deliver · RACI

Agents build fast. I decide what ships.

I built my own harness on the AI-native flow. Every agent reads the same rules, then the work moves from intent to spec, plan, build and review. A shared wiki with git is their memory. I write the intent and I own the gate: nothing ships without my yes.

Agents get the R. I keep the A.

Only I decide: sending, publishing, product calls, accepting work, deleting.

AI at five of six stages. Approved tools only. My sign-off on each.

At work

On my own

Products I run

  • ~30% admin time back
  • 1 h → 15 min call prep
  • −40% time per proposal
  • 2 h → 30 min month-end close

By my own estimates. If a report takes me an hour every week, it becomes a skill.

Approved · Anton

This site, as a project

I did not know Claude could work in Blender. Now there is a 3D room of my real things behind this page. Agents built it, I made every creative call.

  • Oct 26 kickoff
  • 20+ design and review rounds
  • 3 models drawing the same brief blind
  • 10 rough sketches after the reset

A client who comes back. A team that wants to stay.

The best time to be a project manager is now.

AI gives me time back. I spend it on the part only a person can do: asking instead of ordering, working through a disagreement, leading by example.

The new PMBOK Guide puts value first. My version is simpler: the client comes back. Repeat work buys time, and time lets you ask for feedback every cycle and act on it.

Since I was a kid, what I do best is getting people pointed at one goal, in a room where they actually want to stay.

Management was always a creative job. Now I get to do more of it.

Lead and follow. I learned both on the dance floor, long before project management. Ballroom champion of Ukraine, 2010*

Close · Sign-off

Definition of done.

Estimate: 3 min.

Anton Rybalchenko, smiling, walking in a light blue shirt.

antoniorybalchenko@gmail.com · LinkedIn · Resume · New York

Your reading time stays in your browser.

The ten rules I wrote for that team

The client asked for more structure. Before our first retrospective I turned that email into ten rules, and the team followed them from then on.

  1. Monday: release planning and backlog grooming.
  2. Wednesday: status and blockers.
  3. Friday: demo.
  4. A roadmap with milestones, drawn as a Gantt chart.
  5. Every phase cut into parts with a clear scope and goal.
  6. No large tasks added to the current cycle.
  7. Kanban with release milestones.
  8. Tickets drafted with AI from the client's video requests.
  9. A recorded workshop on the CMS, Slack and Jira.
  10. A mini retrospective after every release.

The seven questions I asked

Five design and development leads, July 2025, names removed.

  1. How would you rate the current design and development collaboration on our projects, from 0 (poor) to 10 (excellent)?
  2. What's the biggest friction point in the current design-dev workflow?
  3. Describe a handoff that could have gone better. What happened, and what would have helped?
  4. What do you personally need more of: clarity, documentation, context, flexibility, something else?
  5. What do you wish the other side understood better about your part of the process?
  6. In an ideal scenario, what does design and development collaboration look like on a project?
  7. What ideas, tools or process changes would improve it?

Designers 7.3 · Developers 3.5 · Lowest single score: 1

AG-01

Meeting to Jira

Turns a client call into minutes, ticket updates and RAID entries.

Runs: After every client call · Built with: Claude, Fireflies (no direct connector)

What was the problem?

After every call I rewrote my notes into minutes, tickets and risks by hand.

What did I solve?

The agent reads the transcript and the live Jira board, then proposes minutes, ticket changes, new tickets and risk entries. Nothing is written until I approve.

What was the flow?

  1. Call recorded (Fireflies)
  2. Transcript in, pasted by hand
  3. Meeting agent (skills: Minutes · Ticket changes · RAID candidates)
  4. Proposal in four groups
  5. Human sign-off, group by group, then Jira and Confluence update

What changed?

About 30% of my admin time back, by my own estimate.

AG-02

Discovery brief

Turns a two-line request into a call plan.

Runs: Per new lead · Built with: Claude (company account)

What was the problem?

First calls start from almost nothing, and one missed question costs a week.

What did I solve?

The agent researches the company, its site and its business, then writes a brief with questions by area and a checklist I follow on the call.

What was the flow?

  1. Request comes in
  2. Research agent (skills: Company · Website · Business news)
  3. Call brief and checklist
  4. I run the call
  5. Minutes and next steps, personal data removed in Copilot

What changed?

Call prep went from an hour to 15 minutes. One of my latest clients came in through this flow.

AG-03

Proposal builder

Turns call notes into a reviewed proposal.

Runs: Per RFP or proposal · Built with: Claude (company account)

What was the problem?

Every proposal started from a blank page and went through long review loops.

What did I solve?

The agent analyzes requirements, writes the team brief for the estimate session, drafts the proposal with approach, assumptions, risks and timeline, then checks it against a readiness list.

What was the flow?

  1. Requirements in
  2. Team brief and estimate session
  3. Proposal agent (skills: Draft · Assumptions and risks · Readiness check)
  4. Each lead reviews their part, I set the price
  5. VP sign-off, then it goes to the client

What changed?

About six proposals so far, each in roughly 40% less time.

AG-04

Month-end close

Turns logged hours into budget reports and invoice drafts.

Runs: Monthly, plus a two-week spend report · Built with: Microsoft Copilot

What was the problem?

At month-end I pulled hours project by project and built budget reports and invoices by hand.

What did I solve?

A Copilot skill exports the hours, lays out what the time was spent on and prepares the invoice inputs for each project.

What was the flow?

  1. Month-end
  2. Hours export
  3. Close agent (skills: Budget vs spend · Invoice inputs · Risks and assumptions)
  4. Report per project
  5. I check every invoice, accounting sends

What changed?

Month-end went from about two hours to about 30 minutes, depending on how many projects are active.

AG-05

ForkCast

My iPhone app for eating enough. Photo in, calories and protein on the lock screen.

Runs: Several times a day · Built with: SwiftUI, WidgetKit, an iOS Shortcut with ChatGPT. Specs and reviews by Claude Code, code by Codex.

What was the problem?

I need to eat a lot to gain muscle, and every tracker I tried was built for diets and took too many taps.

What did I solve?

One photo, an AI estimate, a card where I fix the portions, and the day total on my lock screen.

What was the flow?

  1. Photo or text
  2. AI estimate in the Shortcut
  3. Confirmation card, nothing is saved before I confirm
  4. Day totals
  5. Lock-screen widget updates

What changed?

In my daily use, and I started gaining the weight I wanted. 389 automated tests. Next: TestFlight for friends.

AG-08

RunBridge

My iPhone app that hands my runs and recovery to my AI running coach. No more screenshots.

Runs: After every run and every morning · Built with: SwiftUI, HealthKit, the Oura API, iCloud Drive, TestFlight. Managed in Jira in sprints. Specs and reviews by Claude Code, code by Codex.

What was the problem?

I am training for my first marathon with an AI coach. It plans every run from my data, but Apple Health keeps that data on the phone, and my Oura ring keeps sleep and readiness to itself. Every morning I sent screenshots.

What did I solve?

After a run, the app writes the workout, heart rate, sleep and readiness to a folder my coach reads. The main screen shows distance since day one, days to my goal race, the 2027 NYC Marathon, readiness, sleep and the week.

What was the flow?

  1. Run with Apple Watch
  2. The app reads Apple Health and Oura
  3. One file per day in iCloud Drive
  4. The coach reads it in the morning
  5. Today's plan

What changed?

No screenshots since. Built in four sprints. Now a week of real use decides what goes into sprint five.

AG-06

Local PII scrubber

Removes personal data on my laptop before anything goes to a cloud model.

Runs: Every client call I summarize with AI · Built with: Apple on-device models, Python. Claude Code as architect, Codex and Claude Sonnet as developers.

What was the problem?

Cloud models write the best summaries, but a raw transcript sends names, emails and phone numbers to a third party.

What did I solve?

Names, companies, emails, phones and codenames become tags like [PERSON_1]. After the cloud model answers, one command puts the real names back. The lookup table never leaves the laptop.

What was the flow?

  1. Transcript
  2. Scrub on the laptop
  3. Review list, I scan what was left in
  4. Cloud model
  5. Restore real names

What changed?

On a sealed test set it never saw, it caught 175 of 181 personal details. A fresh AI attacker still got 17 of 166 through, so I know where it breaks.

AG-07

LLM council

Five roles argue about my decision, a chair sums it up.

Runs: Before decisions that are hard to undo · Built with: a Claude skill, after Andrej Karpathy's LLM Council idea. Next: Codex and Gemini join as council members, so the roles come from different models.

What was the problem?

One AI answer agrees with me too easily.

What did I solve?

A Skeptic, a Fact-checker, a Visionary, an Outsider and an Executor review the same question separately, then a Chair puts it into one recommendation.

What was the flow?

  1. Question or plan
  2. Five roles, in parallel
  3. Chair synthesis
  4. Risks and next step
  5. I decide

What changed?

It changed real decisions: I moved my food-recognition app to a paid model tier and reworked a design decision.