Issue #66: AI Agents Are Now Your Competition, Gartner's 2028 AI Warning

Good morning.

Most weeks in AI feel like feature launches, pretty demos, and releases that aren’t out yet.

Tons of hype, tons of “Breaking!” and other silly things.

This week, it’s all about what’s here right now.

AI agents are popping up in every business, startup, funding round, and anything else.

Just for a preview, this week, Gartner shifted its language, Citi appointed an AI leadership team, and KPMG launched a multi-agent platform that replaces billable hours with bots.

Agentic AI is moving from speculative demos to deployments everywhere.

Agents are running inside the companies that manage your money, file your taxes, and route your calls.

These aren’t assistants anymore. They’re coworkers.

And the next business advantage won’t come from how well you prompt.

It’ll come from how well you delegate.

Plus, prompt templates to help you figure out workflow automations.

Let’s get into it.

—Sam

IN TODAY’S ISSUE 🤖 

• Gartner: The Agentic Shift Is Real And Permanent

• KPMG Replaces Back-Office Work With Agents

• Nice Launches No-Code Agent Builders

• Citi Scales To 150,000+ Agent-Powered Seats

• Prompt Templates For Workflow Automation

• Interesting Developments You Should Look Into

Let’s get into it.

Gartner Says the Quiet Part Out Loud

Gartner just confirmed what a lot of people building with AI already knew:

By 2028, 30% of what B2B software tools offer today will be replaced. Not by new dashboards or shinier UIs, but by fully-automated AI services that actually do the work for you, end to end.

That’s a big shift.

Instead of using a tool to help you send emails, analyze data, or move a deal forward, it’ll just be done for you. No toggling between apps. No waiting on reports. You’ll give it a goal, and it’ll get the result.

Gartner calls this the move from “LLM development” to “agentic workflows.”

In plain terms: No one cares about building smarter chatbots. You’ll want to build systems that can take actions, make decisions, and follow through without constant human input.

Here’s why this matters for your business:

Most teams still see AI as something that helps around the edges. A faster way to write an email. A tool that summarizes documents. But the real opportunity is doing less yourself.

The companies that win won’t be the ones using AI as a tool.

They’ll be the ones replacing entire tools with AI-powered outcomes.

The new strategy is about rethinking who or what gets work done in your org.

Gartner’s report signals that this is the roadmap most companies are following. Budgets are shifting. Strategies are shifting.

So the smart question now isn’t “how can AI help us?”

It’s: what do we no longer need to do ourselves?

KPMG Replaces the Dashboard With Agents

This week, KPMG launched something called Workbench.

Workbench is a multi-agent AI platform that handles tax, audit, and advisory work from start to finish.

At first glance, it sounds like just another enterprise tool. But it’s not.

This software is doing consultants' job for them while being better and faster.

Agents draft reports. Agents pull records. Agents structure advice. Agents even explain it all in natural language to the client.

And this is coming from KPMG. They’re one of the world’s largest professional services firms. Not a startup chasing hype. 

This is a 100-year-old company telling its clients: “You don’t need to just hire us. You can now hire our software too.”

If you’re running a business today, you’re probably paying for a dozen tools:

One for accounting, one for compliance, one for reporting, one for customer insights.

First off, it’s a lot of subscriptions and bills to track. And second, it usually means you’re bounding between 6-10+ platforms for all your business.

Each one gives you a feature. A dashboard. A task list. Then your team does the work to make sense of it all, even if it means jumping between them all.

Workbench flips that model.

Instead of using software to get partway there, you use agents to get a finished result. It’s like saying “do this for me” to a person with all the software and skills.

That changes the economics of business services.

Businesses no longer need to hire more analysts, add new roles, or grow a team to handle volume. It’s about handing off entire categories of work securely, accurately, and at scale.

What’s important isn’t that KPMG built this.

It’s that they built it for themselves. It’s like they’re actively destroying their own business model so they can build a new one (I call this Bionic Business, a business model made for the age of AI). 

And they’re not alone.

Because once you show clients you can deliver faster results at lower cost with fewer errors, there’s no going back.

So here’s the real takeaway:

If a company like KPMG that was built on human hours, billable rates, and expert labor, is comfortable automating the core of its own services…

What makes you think your business is untouchable?

You’ll need to get out of your own way so outcomes happen faster, cheaper, and with fewer hands involved.

And if your business isn’t already thinking that way?

Workbench is your wake-up call.

NiCE Bring Agents to the Frontline

NICE just launched CXone Mpower. A platform that lets businesses build full-stack, no-code agents to manage customer-facing tasks across the front, mid, and back office.

It’s a major step forward for Agentic AI, because it’s not about automating isolated functions. It’s about building end-to-end workflows that complete themselves.

What happens when a customer reaches out?

The agent identifies their issue, finds the relevant data, takes the right action, and confirms it all in one pass.

No ticket handoffs, bouncing between tools, or need for a human unless the system hits a true edge case.

Let’s break down what that really means for businesses.

Right now, most companies still treat AI like a bolt-on.

• You have a chatbot for FAQ.

• A help desk platform for escalations.

• A CRM that logs history.

• A human team tying it all together when things go sideways.

That patchwork works extremely inefficiently.

It gets expensive. It’s slow to scale. And it makes customers repeat themselves, which kills satisfaction and drains loyalty.

What NICE is saying with Mpower is this: 

You don’t need more layers. You need something that understands the full workflow and owns it.

Mpower agents can:

That last part is huge: memory.

Because what breaks most AI support experiences today is continuity.

Customers come back and have to re-explain themselves. Agents forget what was said. Context disappears between people. That doesn’t just annoy users, it also wrecks your margins.

By giving agents persistent memory and decision rights, NICE is showing us what a modern customer experience could look like when AI actually works like a teammate, not a pop-up.

And the platform isn’t only made for huge enterprises.

Yes, NICE serves big firms, but the Mpower builder is no-code. That means mid-sized companies can start building agents without a dedicated AI team.

You can map out a workflow in plain language. You can define outcomes. You can monitor results and tweak behaviors without writing a single line of code.

That lowers the barrier to entry and brings AI agents closer to product, marketing, CX, and ops teams who’ve historically been blocked from automation by tech constraints.

Now let’s talk about cost.

Support isn’t cheap.

Back-office work like order handling, refunds, and policy updates eats hours without moving top-line metrics. And as customer volume grows, that cost grows linearly.

Traditional thinking says: hire more people, offshore low-cost labor, or try to deflect the load through messaging channels and FAQ.

But agents flip that logic.

You don’t reduce cost by doing less for the customer (at least not directly) you reduce cost by letting AI handle more of the workload accurately, instantly, and at scale.

These agents are all about consistency, uptime, and the ability to act instantly on data.

The long-term play here is obvious: agents become the first layer of execution across the entire customer lifecycle. Not just in support, but in onboarding, renewal, personalization, and retention.

And the companies that start training those agents now will have a massive advantage later, not just in cost savings, but in insight.

Because every time an agent completes a workflow, you gain data. Every decision, every objection, and every preference will be captured, categorized, and ready to shape your roadmap.

So if you’re a founder or operator, here’s the move:

Start thinking about where your company hands off responsibility to tools and teams and whether a well-defined agent could do it better.

Citi Just Gave AI Agents a Seat at the Leadership Table

Citi restructured its leadership to make AI a permanent, strategic focus across the entire company.

Here’s what they did:

• Appointed three senior executives as co-sponsors of enterprise-wide AI adoption

• Committed a portion of their $12B annual tech budget to expanding generative and agentic AI

• Rolled out AI tools to over 150,000 employees globally
  Focused use cases on fraud detection, trade confirmations, and customer service

• Built internal systems like Citi Stylus and Citi Assist, and plugged into Vertex AI for long-term scaling

CEO Jane Fraser called it a transformation priority.

This is a Fortune 100 bank restructuring at the top to bake AI into the bones of their business. If you're running even a small business, there are some very real lessons here.

Let’s start with intent.

Citi is operationalizing.

By putting multiple leaders in charge of AI, not just tech leads, but execs across consumer banking, ops, and enablement, they’re saying AI isn’t a tool for one department. It’s a capability every function is expected to understand and use.

And it’s smart, because the scale they’re targeting doesn’t work without leadership alignment. 

You don’t integrate AI across 150,000 people by adding plugins. 

You won’t integrate AI with your small team of 4 by cobbling together tools here, tools there, and hope for the best.

You do it by building systems, retraining teams, and assigning ownership.

That’s what Citi’s doing.

Now let’s talk about where the Agents come in.

They’re not using AI just for writing or summarizing. They’re experimenting with Agents.

Agents are performing tasks like:

• Automatically flagging fraudulent behavior in real time

• Confirming high-volume trades without manual input

• Managing customer requests across channels using voice and text agents

• Surfacing insights from internal data with minimal human instruction

These systems are live right now. They weren’t talking about what they’re going to do, they told us what they already did.

And here’s the deeper point: when a company this size starts trusting agents to handle high-stakes decisions with real-world consequences, it raises the ceiling for what AI can own.

Because if agents can be trusted with a billion-dollar trade confirmation…

What can they be trusted with in your business?

Citi is creating internal systems that learn from every touchpoint, then feed that intelligence back into their operations automatically.

And it’s a real edge.

You don’t need Citi’s budget for this, but you’ll need their mindset:

• Assign someone real ownership over AI adoption

• Pick workflows with measurable value (time, cost, risk)

• Choose tools that remember and improve, not just react

• Design for delegation, not just assistance

Because once agents prove they can deliver reliable outcomes, they’ll do more than scale work.

They change how your business thinks about who does what.

And if Citi’s betting on that across 150,000 seats?

It might be time to start thinking about how many seats in your company could be handled by agents too.

Prompt Templates for Workflow Automation

These prompts will help you figure out how to automate common workflows. These are three examples but you can modify them to fit your particular needs.

They are best used with a Reasoning Model like o3, Claude 3.7 Sonnet or 4 Opus. Replace brackets with your information and get a fully mapped out and outlined workflow automation.

📱 Marketing Workflow: Content Repurposing & Distribution Pipeline

Your Prompt Template:

My business is a [type of business] that helps [target audience] achieve [main outcome].
We have [team size] people and our main tools are [list all marketing tools you currently use].
I want to automate our content repurposing and distribution process. Currently it works like this:
- Step 1: [Person/role] creates [type of content] in [tool/platform] - takes [time]
- Step 2: [Person/role] extracts key points/quotes in [tool] - takes [time]  
- Step 3: [Person/role] creates social media posts in [tool] - takes [time]
- Step 4: [Person/role] creates graphics/visuals in [tool] - takes [time]
- Step 5: [Person/role] schedules posts on [platforms] using [tool] - takes [time]
- Step 6: [Person/role] sends to email list via [tool] - takes [time]
This happens [X times per week/month] and takes about [total hours] each time.
The biggest problems with this workflow are:
- [e.g., "We often miss key quotes or insights when repurposing"]
- [e.g., "Same person has to log into 5 different platforms"]
- [e.g., "No consistent posting schedule, things get forgotten"]
We need to keep: [e.g., "Final human approval before posting"]
Our budget is roughly: $[amount] per month for automation tools
The content/data in this workflow:
- Comes from: [e.g., "Blog posts in WordPress, videos in YouTube"]
- Gets stored in: [e.g., "Google Drive, Notion database"]
- Needs to go to: [e.g., "LinkedIn, Twitter, Instagram, Email newsletter"]
- Format: [e.g., "Text, images, short video clips"]

💡 Filled Example:

My business is a marketing consultancy that helps B2B coaches achieve thought leadership status.
We have 4 people and our main tools are WordPress, Canva, Hootsuite, ConvertKit, and Google Workspace.
I want to automate our content repurposing and distribution process. Currently it works like this:
- Step 1: Content writer creates blog post in WordPress - takes 3 hours
- Step 2: Marketing assistant reads and extracts 10 key quotes manually - takes 45 minutes
- Step 3: Social media manager writes 15 social posts in Google Docs - takes 2 hours
- Step 4: Designer creates 5 quote graphics in Canva - takes 1.5 hours
- Step 5: VA schedules posts on LinkedIn, Twitter, Facebook using Hootsuite - takes 1 hour
- Step 6: Email marketer adapts content for newsletter in ConvertKit - takes 45 minutes
This happens 2x per week and takes about 9 hours total each time.
The biggest problems with this workflow are:
- Marketing assistant often misses powerful quotes because they skim too quickly
- Designer has to wait for social posts to be written before starting graphics
- VA makes copy-paste errors when moving between platforms
- We have no way to track which repurposed content performs best
We need to keep: Human review of all social posts before scheduling
Our budget is roughly: $300 per month for automation tools
The content/data in this workflow:
- Comes from: Blog posts in WordPress, occasionally YouTube videos
- Gets stored in: Google Drive folders (very messy)
- Needs to go to: LinkedIn (3x daily), Twitter (5x daily), Facebook (2x daily), Weekly newsletter
- Format: Long-form text, pull quotes, square images (1080x1080), some GIFs

💰 Sales Workflow: Lead Response & Qualification Sequence

Your Prompt Template:

My business is a [type of business] that helps [target customer] achieve [main outcome].

We have [team size] people and our main tools are [CRM, email, communication tools].

I want to automate our lead response and qualification process. Currently it works like this:

- Step 1: Lead comes in from [sources] to [where notifications go] - noticed after [time]

- Step 2: [Person/role] manually enters lead info into [CRM/spreadsheet] - takes [time]

- Step 3: [Person/role] sends first response email from [tool] - takes [time]

- Step 4: [Person/role] books discovery call using [method] - takes [time]

- Step 5: [Person/role] sends follow-ups if no response via [tool] - takes [time]

- Step 6: [Person/role] updates lead status in [tool] - takes [time]

This happens [X times per day/week] and takes about [total time] per lead.

The biggest problems with this workflow are:

- [e.g., "Leads wait 2-24 hours for first response"]

- [e.g., "30% of leads never get entered into our CRM"]

- [e.g., "No consistent follow-up sequence, depends who handles it"]

We need to keep: [e.g., "Personal touch in emails, not fully robotic"]

Our budget is roughly: $[amount] per month for automation tools

The lead data in this workflow:

- Comes from: [e.g., "Website forms, Facebook ads, LinkedIn DMs, email"]

- Gets stored in: [e.g., "Should go to HubSpot but often stays in email"]

- Needs to go to: [e.g., "CRM, calendar system, accounting software"]

- Includes: [e.g., "Name, email, company, biggest challenge, budget range"]

💡 Filled Example:

My business is a web design agency that helps local service businesses achieve professional online presence.

We have 5 people and our main tools are Gmail, Pipedrive CRM, Calendly, and Slack.

I want to automate our lead response and qualification process. Currently it works like this:

- Step 1: Lead comes in from website form, Facebook ads, or referral emails to owner's Gmail - noticed after 2-12 hours

- Step 2: Owner copies lead info into Pipedrive manually - takes 5 minutes

- Step 3: Sales person sends personalized first response from Gmail - takes 15 minutes

- Step 4: Sales person sends Calendly link if lead seems qualified - takes 5 minutes

- Step 5: Sales person sends 3 follow-ups manually if no response - takes 30 minutes total

- Step 6: Sales person updates stage in Pipedrive after each interaction - takes 5 minutes

This happens 15-20 times per week and takes about 1 hour per lead when including all follow-ups.

The biggest problems with this workflow are:

- Weekend leads wait until Monday for any response (we lose 50% of these)

- Owner forgets to add leads to CRM when busy (lost 3 big ones last month)

- Follow-up emails are inconsistent - some leads get 1, others get 5

- No automatic disqualification of bad fits (we waste time on $500 budget leads)

We need to keep: Personalized first response that references their specific business

Our budget is roughly: $250 per month for automation tools

The lead data in this workflow:

- Comes from: Website contact form, Facebook Lead Ads, referral emails, occasional LinkedIn messages

- Gets stored in: Should be Pipedrive, but 30% stay in Gmail inbox

- Needs to go to: Pipedrive CRM, Calendly for booking, QuickBooks for proposals

- Includes: Name, email, phone, business name, current website, budget range, timeline, biggest frustration

📦 Products/Service Delivery: Client Onboarding & Setup Process

Your Prompt Template:

My business is a [type of business] that helps [target customer] achieve [main outcome].

We have [team size] people and our main tools are [project management, communication, delivery tools].

I want to automate our client onboarding and setup process. Currently it works like this:

- Step 1: [Person/role] receives signed contract via [method] - takes [time]

- Step 2: [Person/role] creates client folder/workspace in [tool] - takes [time]

- Step 3: [Person/role] sends welcome email with [contents] from [tool] - takes [time]

- Step 4: [Person/role] collects necessary info/assets via [method] - takes [time]

- Step 5: [Person/role] schedules kickoff call using [tool] - takes [time]

- Step 6: [Person/role] creates first deliverable/milestone in [tool] - takes [time]

- Step 7: [Person/role] adds client to communication channels in [tool] - takes [time]

This happens [X times per month] and takes about [total hours] per new client.

The biggest problems with this workflow are:

- [e.g., "Steps get missed - forgot to add last client to Slack"]

- [e.g., "Clients confused about what happens next"]

- [e.g., "Takes 3-5 days before client feels 'onboarded'"]

We need to keep: [e.g., "Personal welcome video from founder"]

Our budget is roughly: $[amount] per month for automation tools

The client data in this workflow:

- Comes from: [e.g., "Signed contracts in HelloSign, intake form in Typeform"]

- Gets stored in: [e.g., "Client folders in Google Drive, project in Asana"]

- Needs to go to: [e.g., "Accounting, project management, communication tools"]

- Includes: [e.g., "Contact info, brand assets, project goals, timeline"]

💡 Filled Example:

My business is a social media management agency that helps ecommerce brands achieve consistent sales through social content.

We have 6 people and our main tools are Asana, Slack, Google Workspace, and Later.

I want to automate our client onboarding and setup process. Currently it works like this:

- Step 1: Operations manager receives signed contract via DocuSign email - takes 30 minutes to process

- Step 2: Ops creates client folder structure in Google Drive (15 subfolders) - takes 45 minutes

- Step 3: Account manager sends welcome email with 5 attachments from Gmail - takes 30 minutes

- Step 4: Account manager sends Typeform for brand assets and passwords - waits 2-5 days for response

- Step 5: Ops schedules kickoff call by emailing back and forth - takes 45 minutes over 2 days

- Step 6: Project manager creates Asana project from template - takes 30 minutes

- Step 7: Ops adds client to Slack channel and shares permissions - takes 20 minutes

This happens 4-6 times per month and takes about 4 hours of active work per new client (plus waiting time).

The biggest problems with this workflow are:

- Folder structure inconsistent (someone always forgets the "Archive" subfolder)

- Clients don't know what happens after signing - they message asking "what's next?"

- Brand asset collection takes forever - we chase them 3-4 times

- By the time kickoff call happens, client enthusiasm has dropped

- No one remembers to add client email to our monthly report list

We need to keep: Personal welcome video from our founder (she records these on Loom)

Our budget is roughly: $400 per month for automation tools

The client data in this workflow:

- Comes from: DocuSign contracts, Typeform responses, email attachments with brand files

- Gets stored in: Google Drive folders, Asana projects, contact info in our Google Sheets "database"

- Needs to go to: Asana for projects, Slack for communication, Later for scheduling, QuickBooks for invoicing

- Includes: Business info, brand guidelines, social media passwords, content pillars, target audience details, past content examples

How to Use These Templates:

1. Choose the workflow that would save your team the most time or frustration

2. Copy the template for your chosen workflow

3. Fill in every [bracket] with your specific information - be as detailed as possible

4. Add any unique steps your business has that aren't in the template

5. Include actual times - even if they're estimates

6. Be honest about problems - the messier the current state, the more impact automation can have

Remember: The more specific and detailed your prompt, the more tailored and immediately useful the automation workflow will be.

Interesting Developments You Should Look Into

💻AI Coding Tools Are Killing the Buy vs Build Debate – Tools like Bolt and Replit are making custom software development so accessible that companies are questioning why they pay for expensive SaaS subscriptions. Who knows how long the traditional software business model will survive? See how SaaS is being disrupted

⚖️Microsoft's Copilot Marketing Claims Under Fire – The Better Business Bureau's watchdog is calling out Microsoft for misleading productivity claims and confusing Copilot branding. This could signal stricter oversight of AI marketing across the industry. Read the regulatory details

💰WhatsApp Unlocks New Revenue for Businesses – Meta is rolling out paid subscriptions and enhanced business features in WhatsApp's Updates tab, creating new monetization opportunities for companies using the platform for customer engagement. Explore the business features

🏢PwC Reorganizes for AI Consulting Boom – PwC is hiring thousands and restructuring its advisory division to meet surging demand for industry-specific AI and digital transformation consulting. Eight new specialized groups are being formed. See the hiring strategy

Agents are acting as team members that show up every day and get things done faster, cheaper, and with fewer mistakes.

That’s not a future thing. It’s a now thing.

From NICE to Citi to KPMG, we’re seeing the same play over and over:

Companies are getting clear about what work still needs a human and what doesn’t.

If you’re still testing AI at the edges, writing blurbs, summarizing notes, playing around in a chat window, this is your sign to stop dipping your toe.

Pick one meaningful part of your business. Map it out and train an agent to own it.

Not assist. Not support. Own it.

Because eventually, the businesses that pull ahead will be the ones doing less work, and still getting better results.

And if you’re building something that matters?

You’re going to need that edge.

Until next time,
Sam Woods
The Editor