Working in a GCC AI team: structure, time zones and what's expected

(Hiring into GCCs is covered in a separate article in this series; this one is about the job once you are in it.)

Three common shapes of a GCC AI team

Many centres mix these. A single GCC can have a platform team that owns its roadmap and an extension team that does not. Ask which one you are joining.

Reporting lines: solid and dotted

In a matrix, you typically have a solid-line manager (appraisal, leave, career) and a dotted-line lead who sets priorities. Friction comes from the gap between them. Two practices help:

• Keep a shared one-page note of your current priorities that both can see. When they conflict, surface it early and ask them to agree, rather than choosing silently.
• Before appraisal season, send your solid-line manager evidence from the dotted-line lead: a short message or review comment about work you shipped. Your manager may not see that work directly.

Time zones decide your calendar

Overlap hours are the main constraint on how collaborative work feels.

Exact overlap depends on daylight saving in the partner country, which changes twice a year. When you receive an offer, ask how many evening calls per week the role expects and whether the team rotates them.

Written work carries more weight than you expect

When decision-makers are abroad, your written documents often are your reputation. Three kinds matter most for AI work:

1. 1Design notes for any model or LLM feature: the problem, options considered, evaluation plan, and risks. Keep it to two pages.
2. 2Evaluation reports with denominators and slices: "correct on 172 of 200 cases; 41 of 50 on the Spanish-language subset". Numbers without denominators get questioned in the next meeting you are not in.
3. 3Handover notes at the end of your day if a colleague abroad will continue the work: what changed, what's blocked, what you need from them.

A short example of a handover note that saves a day's delay:

Handover — retrieval eval, 27 Sep, end of IST day
Done: reran eval on new chunking (400 tokens). Recall@5 up on 31/40 queries,
      down on 3 (IDs in sheet, tab "regressions").
Blocked: need access to the policy PDFs updated last week; request raised.
Need from you: decide whether we accept the 3 regressions for this release.
Next (my morning): add the 3 regressions to the test set either way.

Approval chains for AI work

AI features in large companies often pass through several reviews: security, privacy, legal, model risk or responsible-AI boards, and sometimes works councils or regulators abroad. Reviewers may sit in another country. Expect this and plan for it:

• Find out in your first two weeks which reviews apply to AI features and how long they usually take.
• Prepare the evidence reviewers ask for before you need it — data sources, evaluation results, human oversight, logging. See governance questions before your project is approved.
• Put review time in your estimates. A feature that is built in three weeks but waits five for approval is common.

Moving from "execute" to "own"

The career question in many GCC roles is how to move from delivering tickets to shaping what gets built. What tends to work:

• Bring options, not questions. "Should we use approach A or B?" invites a decision abroad. "I recommend A because of X; B would matter if Y" invites agreement.
• Measure the business result, not just the model. Know what the feature is for and report on it.
• Own something small completely. An evaluation suite, an internal tool, a monitoring dashboard. Ownership of one thing is more visible than contributions to many.
• Build relationships across the time-zone gap. An occasional call at an awkward hour to understand a stakeholder's goals is worth more than many polite status updates.

Questions for interviews and your first month

1. 1Which decisions does the India team make without escalation? Give me a recent example.
2. 2Who sets my priorities day to day, and who writes my appraisal?
3. 3How many calls outside 9 AM–7 PM IST does this role have in a typical week?
4. 4What reviews does an AI feature go through before release, and how long did the last one take?
5. 5Has anyone on this team moved into a role that owns a roadmap? What did they do?

Much of what makes GCC engineers visible — design notes, evaluation reports, stakeholder handling — is taught as enterprise delivery practice in FDE for Professionals, which is still at the proposal stage; its dates and terms are not yet fixed. If you are moving from IT services, the IT services delivery to FDE guide maps the skills you already have.