ChatGPT Agent – Is This the End of NinA AI?

€47,000.

That's what one of our clients at NinA AI Agency saved by automating their customer service. With a solution we built in 3 sprints of 40 hours each.

But here's the plot twist: that same client sent me a screenshot last Friday at 4:43 PM of ChatGPT's new Agent Mode. The agent had just analyzed his Gmail inbox, identified three urgent emails, and drafted responses.

"Olaf, our custom-built agent does this too, right?" Why would we keep this within our N8N environment?"

A fair question.

And honestly, not the first one this week.

After two weeks of intensive testing, I now know why that €47,000 saving would have been impossible with Agent Mode. And why companies switching to "free" agents will pay double in six months.

The agent explosion nobody saw coming

Let me start with what exactly happened. OpenAI's Agent Mode gives ChatGPT the ability to independently choose tools and operate apps like Gmail and Slack. Claude goes even further - you connect even more tools via MCP (Model Context Protocol) like Notion, PayPal and Hubspot.

It may be new to many, but I'm already seeing amazing examples online. Like from the Every team (reading tip!).

Kieran Klaassen and Danny Aziz were debugging agents that created other agents until midnight. Danny tried to build a code-review orchestrator, but ended up with an army of agents with names like "fix-accessibility-touch-targets."

The problem? They existed, but didn't work. "Dude, where are all those files?" he asked via the transcript.

The perfect example of where we are now: the technology is there, but the implementation is still... a work in progress.

N8N Agents vs ChatGPT Agent

Three hours after ChatGPT's Agent Mode launch, I started testing. Not because I had to, but because the question was already in my head: "Can we replace our entire customer service AI Agent workflow with ChatGPT agent?"

The current setup: an n8n workflow that receives WhatsApp messages, consults ERP systems, CRM systems and other business data, and sends automated responses. For customer support, but also for HR.

Average monthly savings for our clients: €7,800. Payback period: 5-6 weeks.

I tested Agent Mode for the same process. Within minutes I had a working Gmail connection (WhatsApp is not (yet?) possible).

Impressive. But then I started digging deeper:

• ▸Verdict 1: Agent only works reactively - no automatic triggers
• ▸Verdict 2: No access to their on-premise ERP
• ▸Verdict 3: Employees each create their own agent prompts = chaos

...and I haven't even mentioned security yet...

A custom workflow with a custom-built AI Agent, combined with fixed AI Automation steps, wins.

Everything is custom-built at delivery speeds 10x faster than traditional software builders.

We're shaking up the market with NinA AI…

The secret of "free" agents

Agent Mode seems free, but here are the hidden costs:

Mid-size company scenario:

• ▸15 employees try it (€375/month)
• ▸5 employees use it actively, 10 forget about it
• ▸Everyone uses their own prompts and everything is still manual, not automated.

Versus our custom solutions:

• ▸Pay per use. Average €50 per client per month in API costs and €20 in hosting
• ▸One-time setup fee and fixed build price
• ▸Average payback period <6 months
• ▸Everyone works with the latest language models and optimal prompts and data.

The difference: real impact on your work processes.

And here's what Big Tech doesn't want you to know: every workflow on their platform creates 3-7 years of vendor lock-in. Look at Zapier's pricing - 518% increase in 9 years.

With us, all workflows and databases become property of our clients. We're happy to manage them, but clients can always take over! I hate lock-ins.

The hybrid strategy works

After all this testing I realized: we're comparing apples to oranges.

Agent Mode = front-end helpers for individuals

Custom workflows = backend orchestrators for business process optimization

The winning combination:

• ▸Level 1: Agents (ChatGPT/Claude) for ad-hoc tasks
• ▸Level 2: N8n workflows for 24/7 business automation

Case study: A SaaS startup combines both:

• ▸Employees use ChatGPT and Claude for research/writing
• ▸Critical processes run on our custom workflows
• ▸Result: 67% total productivity increase vs. 15% with ChatGPT/Claude alone

The MCP opportunity everyone's missing

Here's an insider tip: MCPs (Model Context Protocols) are the technology behind both Agent Mode and Claude Agents. And you can build MCPs into N8N too.

Last month we built MCP connectors for three of our clients.

Result: "Best of both worlds" - n8n workflows with MCP power.

This is our secret weapon for a hybrid solution.

What I now advise my clients

For employees: Start with Agent Mode. Test, learn, experiment. I or my colleague Caroline are happy to help with a talk or workshop!

For teams <10: Start with Agent Mode, but document what works.

For the organization: Focus on workflows and automations from day one. The "quick fix" of prompting yourself will be expensive later.

For everyone: Self-hosting is the answer, secure and GDPR/ISO-compliant.

My vision

Agent Mode is not the enemy. It's an important development.

It makes people comfortable with AI and AI automation. And when they're ready for the next step - for real impact on productivity and ROI, with proper security and a scalable solution - they come to us.

"We're not competing with Agent Mode. We're complementing it."

The future of agents is hybrid. Companies that intelligently combine both have the competitive advantage.

Quick ROI check for your situation

• ▸How many hours/month would automation save?
• ▸What's the average hourly wage?
• ▸If (hours × wage × 12) > €15,000: custom solutions pay off
• ▸Otherwise: Keep experimenting in ChatGPT

Who's in? 🚀 Our September calendar is almost full

Olaf Lemmens
Founder | NinA AI Agency