Self-Improving AI Agents: What Zendesk's Forethought Acquisition Means for Support Teams

The Acquisition Behind Zendesk's AI Overhaul

On March 26, 2026, Zendesk completed its acquisition of Forethought — an agentic AI startup specializing in self-improving customer support agents. This deal is the engine behind Zendesk's May 11 announcement that AI agent capabilities would be included in all plans.

While our previous article covered the pricing implications, the Forethought acquisition raises a deeper question: what does "self-improving AI" actually mean for support teams, and is it the right approach for everyone?

What Forethought's AI Actually Does

Forethought built a multi-agent AI system designed to handle customer support across chat, email, and voice. The core promise is that its AI agents learn from your past tickets and knowledge base content, then generate, adapt, and execute complex workflows — without humans writing rules.

The platform includes several specialized agents: one that resolves customer inquiries autonomously, one that assists human agents in real time, one that routes tickets intelligently, and one that analyzes support interactions to uncover knowledge gaps and optimization opportunities.

The marketing pitch is compelling: AI that gets smarter over time without manual configuration.

The Reality Behind "Self-Improving"

Self-improving AI sounds like a support team's dream. But the reality has important caveats that marketing materials tend to gloss over.

The Data Requirement Problem

Forethought's AI works best with extensive historical data — typically around 20,000 past tickets or more. This is a fundamental architectural requirement, not a nice-to-have. The AI needs large volumes of resolved conversations to learn patterns, identify intents, and generate accurate responses.

For small and mid-sized businesses, this is a non-starter. A 10-person support team handling 200 tickets per month would need over 8 years of history before the AI has enough data to meaningfully self-improve. New companies, startups, and growing teams simply don't have the ticket volume to feed a self-learning system.

"Self-Improving" Still Needs Supervision

Despite the autonomous branding, real-world users report that Forethought's AI requires ongoing manual oversight. Teams need to constantly monitor outputs, tweak intent classifications, and review automated responses for accuracy. CSAT analysis remains largely manual rather than AI-driven.

Self-improving doesn't mean set-it-and-forget-it. It means the AI can suggest optimizations — but someone still needs to review, approve, and course-correct.

Data Lock-In Risk

When an AI learns from your tickets, it creates a form of vendor lock-in that goes beyond typical SaaS switching costs. The model's understanding of your customers, your products, and your support patterns becomes embedded in the vendor's system. Migrating away means starting over with a new AI that knows nothing about your business.

This is a strategic consideration that often gets overlooked in the excitement around advanced AI capabilities.

Enterprise AI vs. SMB Reality

Forethought was built for enterprise support operations — large teams with massive ticket volumes, dedicated AI operations staff, and custom-priced contracts. Zendesk acquiring this technology and rolling it into all plans doesn't change the underlying requirements.

For teams under 50 agents, the key questions are:

• Do you have enough data? — Self-improving AI needs thousands of resolved tickets to learn from. If you're handling under 1,000 tickets per month, rule-based and knowledge-powered AI will likely outperform a data-hungry self-learning system.
• Do you have AI operations capacity? — Someone needs to monitor, tune, and validate AI outputs. Enterprise teams have dedicated roles for this. SMBs usually don't.
• Is unpredictable billing acceptable? — As AI handles more interactions autonomously, per-resolution fees become a larger and less predictable portion of your support costs.

The Case for Human-Guided AI

There's an alternative approach that works better for small and mid-sized teams: human-guided AI that draws from curated knowledge rather than trying to learn autonomously from raw ticket data.

This is the approach DeskLeap takes. Instead of requiring tens of thousands of historical tickets, our AI works from day one because it's powered by your knowledge base — the articles, guides, and documentation your team has already written.

Here's how the two approaches compare:

Aspect	Self-Improving AI (Forethought/Zendesk)	Knowledge-Guided AI (DeskLeap)
Data requirement	20,000+ historical tickets	Works with your existing knowledge base
Setup time	Weeks to months	Minutes — toggle on in settings
Ongoing maintenance	Dedicated AI ops team	Update your KB articles as usual
Accuracy source	Learns from past interactions	Draws from curated, verified content
Best for	Enterprise (50+ agents, high volume)	SMB and growing teams (any size)
AI billing	Per-resolution fees; rate not published	Included with the AI Agent, no usage fees
Vendor lock-in	High (model trained on your data)	Low (your KB is portable)

What Support Teams Should Do Now

The Zendesk-Forethought combination will be a strong offering for large enterprises. But for the majority of support teams, here's what matters:

1. Invest in your knowledge base first. Whether you use DeskLeap, Zendesk, or any other platform, a well-organized KB is the foundation for effective AI. Self-improving or knowledge-guided, every AI performs better with quality content to reference.
2. Match the AI to your scale. A 15-person team doesn't need enterprise-grade self-improving agents. You need AI that handles common questions accurately, routes complex issues to humans, and doesn't surprise you with variable billing.
3. Watch for Zendesk Relate announcements. The May 18–20 conference will likely reveal how Forethought's technology is being integrated and what it means for pricing. Wait for specifics before making platform decisions.
4. Evaluate total cost of ownership. Factor in AI operations overhead, per-resolution fees at your expected volume, and the cost of migration if you decide to switch later.

The Bottom Line

Zendesk acquiring Forethought is a significant move that signals where enterprise support is heading. Self-improving AI agents are real technology with real capabilities — for organizations with the data volume and operational maturity to support them.

For small and mid-sized teams, the better path is AI that works with what you have: a solid knowledge base, configurable automation modes, and predictable pricing. That's what DeskLeap delivers — AI-powered support that's effective from day one, without requiring 20,000 historical tickets or a dedicated AI operations team.

Try DeskLeap free and see how knowledge-guided AI can transform your support workflow — at any scale.