Forward-deployed AI consultant · Embedded AI implementation for scale-ups · Lisbon, on-site across Europe
I work inside your teams to redesign one costly workflow, deploy it with your people, and measure the net benefit with your COO and CTO.
Then I set up ownership so it keeps running after I leave.
This is one of nine tools we put into production in six weeks. Here is who runs it today.
IMPLEMENTATION TEAM · MAINTAINED IN-HOUSE SINCE 2026Inside a global HR and payroll platform, deployed 2026. Used daily by course owners, new hires and implementation clients. Maintained by the client's own team.
Report W1: what was built and who runs itAccounts with certified users against accounts without, six-month window, 5,000+ people certified. Correlation, not a controlled experiment. Cohorts and method in the report.
Report W3: what the cohort data showedTwo learning platforms into one after an acquisition, 2022 to 2024. Content production cost down 50%. Annualised run-rate and realised savings shown separately, with the baseline.
Report W5: baseline and what was realisedWho this is for
How I work · 12 weeks, 4 stages
Each stage ends with something you can see and sign off. Payment follows acceptance, not the calendar.
I sit with the teams, map the workflow from real tickets and documents, collect volumes, handling time, errors and the automations you already have. Data, access and security are checked with your CTO.
Prototype with the team, port to production on your infrastructure, tests, security review. Code lives in your repository from day one.
A named owner in the team, training on real work rather than slides, measurement against the baseline.
Documentation, who maintains what, a plan for the next workflows, and a checkpoint 90 days later.
First workflow milestone
Counted from access and security approval. No promise about your security team's decision or the business result. I guarantee delivery and measurement.
Deployment reports
Written for the person who has to defend the spend to a board. Each report says what I built, who else delivered, what it cost to run, and what I would not repeat.
Course authoring replacing a licensed tool, a new-hire portal, a client implementation hub with 197 checks, a video studio. Which were AI, which were plain software, and the maintenance cost against the licence they replaced.
5,000+ people certified in six months. Tickets per account and product adoption, certified accounts against the rest. What was customer education and what was AI automation, kept apart. Programme cost and whether the effect held.
LMS operating cost down 70%, content production cost down 50%. Annualised run-rate separated from realised savings, with the baseline, the period, and what the migration cost.
Verifiable today: the agent pipeline and video pipeline behind W1 are documented in a public repository, lackiantoni.github.io/learning-design-samples. Technical evidence, not business evidence.
Ways to work together
Not a ladder. The Assessment can end with "do not deploy", and you keep the report.
Economic baseline of one to three workflows, ranked opportunities, feasibility reviewed with your CTO, data and security dependencies, the scope of a first deployment.
One department, one contracted workflow in production with acceptance criteria. Further workflows depending on complexity. Owner in the team, measurement against baseline, handover.
Weekly steering with the sponsor, a monthly board report, decisions on the workflow portfolio, model and usage rules, and acceptance criteria for new deployments.
Three defined ways of working with AI on real tasks, an owner per team, usage measured before and after. Not workshops.
No new licences by default. Most companies already pay for more AI than they use. If something is genuinely missing, the Assessment says so with the cost.
IP is yours. Code sits in your repository from day one. Thirty-day exit clause. No subcontracting without your consent.
Operating brief
The full brief covers data and security, the CTO relationship, boundaries, continuity and availability. Short version here.
I work in your accounts and your tenants, with the model providers you approve, no subcontractors, NDA as standard, and a security review before anything reaches production.
Your CTO or their delegate approves the architecture, allocates a few engineering hours a week, and owns maintenance after handover with documentation I write.
No company-wide strategy without a deployment, no customer-facing chatbots, no replacing your development team, no 24/7 support.
IP is yours and code is in your repository from day one. Thirty-day exit clause. Documentation plus a recommended partner cover absence.
Three days a week on your deployment, one week a month on site. Based in Lisbon; London, Warsaw and Berlin regularly. Next start date on the call.
First-workflow hypothesis
No email needed to see the result. You get one suggested workflow, the assumptions behind it, what we do not know yet, who should own it, the matching model with its price, and the agenda for a 30-minute call.
About
I have built three academies from zero, for employees, partners and customers, and I know what happens to a tool after the consultant leaves. That is why ownership is part of every scope.
Since 2024 I build and ship internal tools by directing coding agents (Claude Code, Codex), with the client's engineers reviewing architecture and security. Before that: customer education after an acquisition in translation software, learning design at a fintech, a bank and a travel-claims company. Google AI Professional Certificate.
I live in Lisbon and work on site across Europe.
More about me and the companies I have worked withNotes, every two weeks
What the team adopted, what they quietly stopped using, what it cost. No course pitch. Unsubscribe in one click.
Questions before the call
I sit with the team that does the work, in your Slack, your tools and your meetings, for the weeks of the engagement. I map the process from real tickets and real documents, not from a workshop. The tool is built in your repository, on your infrastructure, and the people who will use it see it from the first prototype.
Your CTO, or the person they name, approves the architecture before anything touches production. Security review happens before launch, not after. I need a few hours a week from one of your engineers for access, code review and handover. Conflicts go to the executive sponsor.
Usually not. Most companies I meet already pay for more AI than they use. If a model or a tool is genuinely missing, the Assessment says so with the cost, and you decide.
Two weeks. A baseline of one to three workflows: volume, handling time, errors, existing automation. Feasibility reviewed with your CTO. Data and security dependencies. A scoped first deployment with acceptance criteria. And a verdict: go, plain automation, simplify the process, or do not deploy. "Do not deploy" is a result you paid for, not a failed sale.
By week 12 a named person in your team owns the workflow, with documentation, a maintenance owner and a plan for the next workflows. We check in at 90 days. The code has been in your repository since day one, so nothing leaves with me.
The Program is three days a week on your deployment. Fractional leadership is two contracted days a week with a reply within one working day. Both are written into the agreement.
Based in Lisbon. On site one week a month, usually London, Warsaw or Berlin; other European cities by agreement. The rest remote, in your time zone.
Company-wide AI strategy without a deployment attached. Customer-facing chatbots. Replacing your development team. Round-the-clock production support after handover. If you need one of those, I will say so on the first call and, where I can, name someone who does it.
Three fields and the workflow you have in mind. I reply personally with two time slots.
Your first-workflow hypothesis
A hypothesis, not a diagnosis. Six answers cannot establish data readiness. The call does.
Book a 30-minute call
Three fields. Before the call I will only ask what company you are, what your role is, and which workflow you have in mind.
Prefer email? kontakt@antonilacki.com