Your AI Shouldn’t Need a Prompt for Every Task: Gemini Enterprise and Antigravity 2.0

Abel Nyamekye

AI Learning Academy · Session 11 Recap

Session date: September 16, 2026Presented by: Wursta AI Center of ExcellenceFormat: Live webinar + demo

Most organizations met generative AI in a side panel: summarize this document, draft that reply. It is useful, but a person still has to start every task. In Session 11 of the AI Learning Academy, Wursta cleared up what has changed inside Gemini Enterprise and then demonstrated Google Antigravity 2.0 doing real work live: auditing Google Drive, assembling status reports, and completing a timesheet in a system that has no API.

60%of poll respondents want to build multi-step workflows without code
44%named security and governance as their top AI blocker
2–3 hrssaved per day, by our principal AI architect’s estimate
~$40compute cost of a comprehensive Drive analysis in our demo

From AI that helps you to agents that work for you

We opened with a familiar scene from Ferris Bueller’s Day Off: a homemade contraption of pulleys, a tape recorder and a dummy that handles several jobs at once while its builder is elsewhere. That is close to how knowledge work runs today. People move between mail, documents, spreadsheets and third-party applications to finish one process. The shift we presented is from AI that waits for a prompt to agents that carry a goal through several steps on their own, inside limits the organization sets.

Sidebar AI

  • Starts when a person types a prompt
  • One task, one app at a time
  • Stops after each answer
  • Personal productivity

Governed agents

  • Starts from a goal or a schedule
  • Works across apps, files and the web
  • Keeps going until the job is done
  • Team execution under IT controls

Ferris, I would call an early adopter of automation. Terrible ethics, of course, but great instincts.

Christopher Hopkins-Ward, AI Enablement Lead, Wursta

What the audience told us

Attendees included IT directors, system administrators, technology vice presidents and operations leaders. We ran three live polls during the session, and the answers shaped the rest of the agenda.

What’s the biggest thing slowing AI down at your org?

9 respondents · multiple answers allowed

Security and governance questions44%
Cost and budget predictability33%
Making it work with existing tools22%
Getting people to actually use it11%

Which of these would help your team most right now?

10 respondents

Multi-step workflows without code60%
Connecting the tools we already use20%
Fast spreadsheet and document analysis10%
Compliance-grade security (FedRAMP High)10%

How ready are you to let an agent work on its own?

3 respondents · multiple answers allowed

Nervous about permissions and security
Show me real use cases first
Ready, if we control the guardrails
Just here to see what’s possible

Share of respondents to each poll, rounded. Source: Wursta AI Learning Academy Session 11 live polls, September 16, 2026.

The pattern is clear. The main obstacles are control and cost rather than capability: security, governance and budget predictability accounted for most answers to the first poll. What teams want most is the ability to build multi-step workflows without writing code. On readiness, the sample was small, but no respondent was prepared to hand over autonomy without conditions. Every answer asked for guardrails or proof first.

Gemini for Workspace: four updates worth knowing

Before turning to Gemini Enterprise, we reviewed recent Gemini for Workspace changes that apply to every Workspace customer. All four are documented on the Google Workspace Updates blog.

Third-party connectors

Workspace Marketplace integrations can connect to Gemini once an admin approves them. Rolling out to Rapid and Scheduled release domains first.

External notebook sharing

Gemini notebooks were internal-only. Admins can now manage external sharing, so clients and partners can collaborate in them.

AI data classification

In open beta for Drive. Admins write labeling instructions and Gemini applies them, replacing manual training sets. Labels can drive DLP and retention.

Instructions everywhere

Saved personalization now applies across the side panel in Gmail, Drive, Docs, Sheets and Slides. Set “show me action items first” once.

Clearing the fog: what’s new in Gemini Enterprise

Organizations that last evaluated Gemini Enterprise a few months ago will find a different product. The interface and the agent tooling have both changed, and we walked through the changes live.

Workflow Builder

Agent Designer is now Workflow Builder. Chain steps, run them on a schedule, pick the model and output format, connect apps, and share workflows as internal templates.

Demoed live

Skills

Package a repeatable procedure once and reuse it. Create skills by describing them, install Google’s (contract review, brand voice and more), or upload your own.

Demoed live

FedRAMP High data stores

Added at the end of August, opening the door for government, healthcare and finance teams with strict compliance needs.

Gemini 3.8 Flash and Slack

A new model benchmarking well on finance and data-reading tasks at a low cost per capability, plus a revamped Slack connector from early September.

Instead of explaining the same procedure to an AI every time, you teach it once, and then you package it.

Christopher Hopkins-Ward, AI Enablement Lead, Wursta

How Gemini Enterprise is priced

Business$21 / seat

Teams of up to roughly 300 with little IT lift. Does not include Antigravity.

Needed for Antigravity$30 / seat

Standard. Unlimited seats, higher quotas, third-party agents and a stronger compliance story.

Plus & FrontlineCustom

Priced through sales; Google does not publish these rates. Plus also supports Antigravity.

Pay as you go$0 base fee

Pay only for what you consume. Useful for proving value before a commitment.

Starting prices as presented on September 16, 2026. Heavy use consumes compute that is billed separately from seats, so budget for both. Confirm current pricing with your Wursta account manager.

What Antigravity 2.0 is, and what it can reach

The Antigravity name is also attached to Google’s coding tools. Antigravity 2.0 is different: a desktop application for macOS, Windows and Linux, built for office work, with agents that carry out multi-step tasks such as weekly status reports or turning a spreadsheet into a deck. Work can run on a schedule, so drafts are waiting when the user arrives. Because it runs on the user’s machine rather than in the cloud, scheduled work only runs while that machine is on.

Antigravity 2.0Runs on the user’s own Mac, Windows or Linux machine

Google apps

Gmail, Docs, Sheets, Slides, Drive, Vids and Chat. Connectors sign in as the user, credentials stay on the laptop, and the agent sees only what the user can see.

Local files

Works with files on the computer, limited to the folders the user defines before anything runs.

The web

Drives Chrome in its own separate profile, with an allow list that starts empty. This is how agents work in systems that have no API.

Governance: high autonomy, central control

With security and governance at the top of the first poll, this was the part of the session that drew the most attention. An agent that can operate a computer and drive a browser raises a fair question about misuse. Administrators get visibility into how Antigravity is used and what it costs through the Admin console and Google Cloud console, along with these controls:

Approved tools onlyAdmins choose which MCP servers and models are allowed. No unvetted data stores.
Scoped accessBrowser allow list starts empty. Loading a page and clicking on it are separate permissions.
Spend limitsHard budgets per project, with lockdown when a team goes over.
Full audit trailPrompts, responses and metadata are logged in Google Cloud.
Human sendsAgents draft. The send action isn’t granted, so a person reviews everything first.

It’s high autonomy, but there’s central control over that. It’s like a dial that you can turn up to eleven if you want to, but you have that control.

Christopher Hopkins-Ward, AI Enablement Lead, Wursta

Later in the session we showed these controls in a live console: browser-access settings, budgets in the Google Cloud billing console (billing-viewer access is required), the approved-model list, and usage and spend metrics broken down by person and by project.

Live demo: Antigravity at work

Rather than a staged scenario, Wursta’s principal AI architect walked through the agents he runs every day in his own role. Each one started as a plain-language goal.

Weekly competitive analysis

In goal mode, agents research Microsoft Azure, Google Cloud, AWS Bedrock, Atlassian and Oracle from the technical foundations up, then compare them. About an hour of unattended work per run.

Goal mode

Always-current project report

Scans local and shared folders, classifies each project as internal or external, dates it by activity, and uploads the report to Drive, where Gemini Enterprise can answer questions about it.

Scheduled

Drive audit, turned into a skill

A tiered audit reads folder structure, then file metadata, then content. A comprehensive run cost about $40 in Vertex AI. The process was then packaged as a skill for Gemini Enterprise.

Skill

Timesheets with no API

Browser automation logs into the timesheet system, maps a week of Workspace activity to the right project codes, and fills it in, saving a screenshot and logs as proof.

Browser automation

Here’s the goal. Do the work. Figure out how to do it until you’re done.

Jacob Ortony, Principal AI Architect, Wursta, describing Antigravity’s goal mode

Inside the timesheet automation

The timesheet demo addressed the problem many attendees raised: important systems that have no modern API and would otherwise require a costly custom integration. The agent works through the same screens a person would.

Open the siteLaunches the timesheet in its own browser profile.
Read the optionsCollects the projects available as toggles and fields.
Research the weekReviews Workspace activity day by day: documents, Gemini meeting notes and more.
Map the workMatches tasks and documents to the right projects.
Fill and recordCompletes the entries and saves a screenshot, logs and a data breakdown.

We also explained why running on a laptop matters. Cloud agents are built for efficiency: they start with a small, locked-down set of tools, complete one job and shut down. A desktop agent can use whatever tools the organization allows. In our demo, a Chromebook ran about 20 agents at once, and the same setup can diagnose a slow Wi-If connection or a meeting-audio problem from the system logs and explain the fix in plain language.

Typically, people only use about 5% of the capabilities of their laptops. On a Chromebook, I might have 20 or so agents running simultaneously.

Jacob Ortony, Principal AI Architect, Wursta

Six workflows to start with

For teams deciding where Antigravity fits, we recommended starting with recurring work that consumes hours every week and already follows a pattern.

Pre-meeting briefsTwenty minutes of searching threads and notesA one-page brief ready before the call
Meeting follow-upsRecaps written from memory after the factDrafted recap and next steps to review and send
Weekly status reportsRebuilt from scratch every FridayCompiled in the background from the week’s activity
Inbox triageManual sorting before real work startsPrioritized inbox with draft replies waiting
Receivable remindersTracked by hand in a spreadsheetReminder drafts prepared on a schedule
Deck from a spreadsheetCopying numbers slide by slideA locked Drive template filled with the latest data

How to get started

Organization-wide use with central governance requires a Gemini Enterprise Standard or Plus seat. Individuals who want to explore first have a lower-cost path:

  • Antigravity 2.0, the desktop app shown in the session, is a free download and can be explored with a personal Gmail account.
  • Antigravity CLI is similarly capable and better suited to developers and engineers.
  • Antigravity IDE is not yet integrated into the wider ecosystem, so we recommend holding off on it for now.

These tools were for developers, but they’re emerging as the most powerful extensions of Google Cloud and AI in the workplace. I highly recommend learning this.

Jacob Ortony, Principal AI Architect, Wursta

How Wursta helps

AI Center of Excellence

Compass Essentials includes AI Learning Academy access, monthly live office hours and quarterly AI leadership sessions. Compass Standard and Compass Visionary add company-wide training, the Wursta adoption agent and dedicated time with our engineers.

Deployment Kickstart

Gemini Enterprise packages for smaller organizations, with ready-made agent libraries and video training. Upgrades add cloud platform setup, secure foundations and additional system connectors.

Ready to put a governed agent to work?

Watch the full session, then talk with your Wursta account manager about a secure Gemini Enterprise and Antigravity rollout for your team.

The Wursta AI Learning Academy runs live on the third Wednesday of every month. To suggest a topic, contact your Wursta account manager. Session 10, on Workspace Studio, is available on demand.

Stay up to date

Stay informed with expert guidance on cloud strategy, digital transformation, and the evolving tech landscape—delivered straight to your inbox.