Configure what you already pay for

Most companies already pay for AI seats and use them as bought: each person pastes in what they need and checks the answer alone. Level 2 turns the same seats into a company AI workspace. No new licences, four pieces of work, and it is the level most companies skip.

Dušan StamenkovićFounding & Managing Partner · AI Practice LeadPublished 19 October 2026 · 2 min read · Cross-industry · Level 2 in practice

01

The four pieces of work

WorkWhat it meansDone when
ConnectionsThe assistant reads from where documents actually live: file shares, the wiki, key systems, through the vendor's connectorsThe sources people ask about most are connected
PermissionsThe assistant sees only what each user is allowed to seeA permission review has passed, and oversharing is fixed
Onboarding by roleReady-made projects, prompts or assistants for sales, legal, quality, financeEach non-technical role has several workflows of its own that work
One owner per knowledge domainSomeone keeps each area's documents current and retires old versionsEvery connected area has a named owner

02

Fix oversharing before you connect

Once a source is connected, the assistant can surface any file a user can technically open, including the ones nobody remembers sharing. Review who can see what first. It is the least visible step, and the one most likely to cause an incident.

03

What to measure

—

Weekly active users by role,not seats bought.

—

Questions answered from connected sources,against answers from the open web.

—

Time saved in a few named workflows,sampled, not surveyed. Pick several workflows from the non-technical teams and time them before the rollout: that is the baseline. Engineers are a separate case; they work with skills and agent harnesses on their Claude subscriptions.

—

Stale documents retiredper owner each month.

04

When Level 2 is not enough

Move up when the answer has to combine several systems, act on records, or reach customers directly. Until then, configuring what you already pay for is usually the cheapest AI investment on the list.

On the site: the use-case page Enterprise AI search, and the pillar, The four levels of building AI systems.

Practice Notes · for CTOs and tech leadsMore in Insights →

Ask how this applies to your own stack.

Schedule a call with a partnerFree until 11 December