Do you need access to our source code or repositories?

No. No system access is required. Permissions are not granted and the repository is never opened. We examine checks and delivery workflows as operational objects. If agreed in advance, the process owner may share their screen to show branch protection settings or build logs, but we never touch your codebase.

What happens when no reliable check exists in our repository?

That is a standard and legitimate finding. The session will conclude with either Outcome 2 (autonomy with mandatory human diff reading, where building the check becomes the pilot scope) or Outcome 3 (do not automate). You avoid funding automation that cannot safely land changes.

How does this differ from your practical AI workshop for operators?

The practical AI workshop for operators is designed for business operators (owners, CFOs, COOs, VP Operations) addressing margin, inventory, ERP systems, and business process automation. This session is built strictly for engineering leaders, CTOs, and delivery heads, focusing on software delivery pipelines, repository checks, and cost per completed change.

What is a negative control and why is it required?

A negative control is an intentional breakage of the target behavior to prove that your automated check actually goes red. Without a negative control, a passing check is only a hypothesis. Client engineers run this experiment within a week after the session before any autonomy is permitted.

What follows the two-hour session?

You receive the complete package within two business days. If an autonomous candidate is verified, you can fund a scoped 4 to 6 week pilot, build it with your internal team, or decide not to automate.

What language is the session conducted in?

The session runs in English or Russian, matching the preference of your process owner as agreed on the intro call. Deliverable documents default to English, with Russian available upon request.

How are our data and intellectual property protected?

We execute a mutual NDA before the intro call. Because we require zero repository or system access, your proprietary code never leaves your perimeter. All deliverables and pilot plans belong entirely to you.

Free 2-hour workshop · AI Factory Pilot Design

Lower the cost of the work your team repeats.

Q: Who should attend the pilot-design session?

There is a strict ceiling of no more than four people from your team. The process owner who personally landed a change in this pipeline within the past month is mandatory. The engineer who most recently edited the checks or build pipeline is strongly recommended.

Two hours, online, free, for engineering leaders. One workflow. One pilot plan. No system access required. We take one recurring delivery workflow, design it, and decide whether a pilot is worth funding.

2 hours·Free·Online·CTO-led·No repository access

Book an intro call See how it runs

Two hours, one sitting. Free, and the bar stays the same: four people from your side at most, and the process owner who personally landed one of these changes in the last month has to be one of them. Without that person in the room the session runs on guesses.

From session to a funded pilot

Your recurring delivery task

Ticket intakeCode changesReview and rework

2-Hour Pilot Design

Cost per completed changeHuman-agent boundaryNegative control card

Engineering outcomes

Autonomy candidateOr mandatory diff readingOr do not automate

Design session

Ground automation in code delivery economics

This is a pilot-design session, not training or a production build. Throughout the two hours, one number stays on the screen: cost per completed change. The process owner breaks it down out loud. The stream map explains what the number consists of, the human-agent boundary defines what may be touched, and the result check decides whether we have the right to touch it without a human reading the diff.

What the session is not

Not training and not a tools overview
Not a repository maturity audit
Not building automation
Not a promise of savings: before the pilot, all numbers are the process owner's estimates

What you leave with

One recurring delivery workflow focused on cost per completed change
A verified boundary between inner execution and human decisions
A designed negative control experiment to prove check validity
A scoped pilot plan and brief delivered in two business days

Have one recurring delivery workflow that consumes team hours? Let us look at it.

Book an intro call

Unit of account

Cost per completed change

We measure the cost per completed change for one task class, not abstract engineering efficiency. Throughout the session, this number stays on the screen. The process owner breaks it down across five concrete components.

Author active work

Direct engineering time spent writing the change.

Review as separate hours

Time spent by reviewers reading diffs and evaluating code.

Rework and returns

Rework rate and engineering hours per returned pull request.

Coordination and context re-entry

Context switching, meetings, and ticket coordination time.

Tooling cost per change

Direct infrastructure and tool costs, including existing tool subscriptions.

Session decisions

Three valid outcomes

All three outcomes are legitimate. Leaving with a clear conclusion that a task should not be automated protects your engineering budget and stops unviable projects before they start.

OUTCOME 01

Candidate for autonomy found

A candidate for autonomy is found, verified by negative control on client code. Scoped into a pilot plan to lower the cost per completed change.

Autonomy candidate

OUTCOME 02

Autonomy with mandatory diff reading

Automated assistance with human review required. The scope of the pilot becomes building a reliable, tamper-resistant check.

Check building

OUTCOME 03

Do not automate

No check with the required properties exists, or volume does not justify investment. A clear stop that prevents wasted engineering budget.

Legitimate result

How it runs

The two-hour session agenda

Online format. One document on screen throughout the session, no slides. The process owner types into the shared structure while we guide, challenge, and analyze the delivery boundary.

The unit of account

Align on measuring cost per completed change for one task class, establish session rules, and set expectations.

Foundations and rules

Five cost components and volume

The process owner breaks down author work, review, rework, coordination, and tooling costs, multiplied by monthly frequency.

Current cost breakdown

Stream map tied to cost

Map the six steps from ticket intake to accepted result, identifying who decides, where rework occurs, and where wait time sits.

Workflow bottlenecks

Human-agent boundary

Separate the inner execution loop from outer human decisions. Assemble the evidence bundle required to justify each step.

Guardrails and diff reading

Result checks and control card

Examine candidate checks for tamper resistance. Design a negative control experiment that client engineers run after the session.

Check verification

Risk gates and pilot brief

Evaluate landing authority, deploy separation, and rollback. Write a scoped pilot brief with acceptance criteria and week-one baselines.

Pilot decision

Preparation

What you bring to the session

One workflow. One pilot plan. No system access required. We receive no repository access, permissions are not granted, and your codebase is never opened. You bring your numbers and your team.

Sanitized ticket example

One real ticket in the customer's original phrasing, with proprietary details and secrets removed.

The process owner

The engineer or lead who personally landed a change in this pipeline within the past month. Mandatory.

Frequency and hours estimates

Realistic monthly task frequency and estimates of active engineering hours spent per change.

Output consumer definition

Clear answer to what the consumer sees and who will notice if an output or number turns out to be wrong.

Four-person room ceiling

Strict ceiling of no more than four people from your team to keep math and decisions focused.

Deliverables

What you take away

A single document delivered two business days after the session. Six distinct sections assembled directly from session findings.

Economics sheet

Current cost per completed change across five components, task volume, calendar time, proposed delta, and named savings recipient.

Workflow and guardrails

Stream map marked with decision points, rework loops, wait times, boundary definitions, and four risk gates.

Result checks registry

Candidate checks evaluated against failure modes, with status and verification requirements.

Negative control card

Execution-ready experiment card for client engineers with designated assignee, return date, and deliberate failure test.

Scoped pilot brief

Single bounded task, acceptance criteria, named owners, four metrics, and week-one baseline measurement plan.

Not checked inventory

Itemized inventory of unverified items, empty gate cells, and forecast numbers with documented reasons.

Want an objective decision on whether a software delivery pilot is worth funding?

Book an intro call

Positioning

Built for engineering leaders, not business operators

Who this session is for

Engineering leaders, CTOs, heads of delivery, and technical practice leads at software teams, IT services firms, and their delivery partners.

Teams with recurring delivery workflows running at least weekly
Leaders looking to lower cost per completed change with verified checks
Teams seeking to define human-agent boundaries on code pipelines
Engineering organizations deciding whether a pilot is worth funding

Book an intro call

Who this session is not for

This session is not for business operators, CFOs, COOs, or business owners looking to apply AI to business operations like inventory, ERP workflows, margin leakage, or demand forecasting. It does not cover business workflows outside the software delivery pipeline.

Looking for operational business workflows?

If your focus is business operations, distributor margin, or ERP integration, our practical AI workshop for operators is designed specifically for your leadership team.

Explore the practical AI workshop for operators

Honesty

What we deliberately do not promise

Stated out loud during the session and recorded as a dedicated section in your deliverable package. Clear boundaries, zero hype.

Permission for autonomy

The session delivers a candidate and an experiment design. The right to run unattended is earned only after client engineers execute the negative control.

A verified repository maturity audit

Without code access, repository state is an unverified estimate based on interview answers, not a formal audit.

Measured savings before a pilot

All session numbers are process-owner estimates and forecasts. Real baseline measurement begins during week one of a funded pilot.

A borrowed change size ceiling

Safe change size ceilings must be calculated from your repository's own 75th percentile merge history, not borrowed benchmarks.

Tool comparisons and agent selection

The session designs the workflow, result checks, and human-agent boundaries. Tools and models do not alter the flow map.

An evaluation suite

Building a comprehensive eval harness requires repository access and 20 to 50 real tasks, which belongs in a pilot build.

Critique of existing team checks

Checks are analyzed strictly as operational objects, without evaluating author or team performance.

Why us

Built to deliver, not to hype

Leadership

CTO-led

Vladimir Gorshunov: 20+ years in delivery, fluent with C-level.

Partner

Anthropic Partner

Member of the Anthropic Partner program. Certified engineers.

Track record

20+ in production

HIPAA de-identification, banking call quality, wellbeing AI (4.8 stars).

Team

Senior engineers

Engineers who ship production software and know how checks fail.

See more of the work in our case studies or explore our operational solutions.

FAQ

Questions engineering leaders ask

Get in touch

See if the workshop fits your team

Tell us which delivery task repeats and who owns it. We will say whether the workshop is worth two hours of your team's time, or why it is not.

Email us directly

sales@3alica.com