What Freshworks announced
At Refresh 2026, Freshworks did not just ship features, it placed three strategic bets about how B2B AI will be built, sold, and measured: the Freddy AI Agent Studio, an MCP Gateway for open interoperability, and Experience Level Agreements (xLAs) that tie service performance to employee sentiment. As an engineer who spent four years inside Freshworks’ platform, I read these differently than the launch posts want you to.
Bet 1: Open standards (MCP Gateway)
Freshworks chose interoperability over lock-in at the exact moment Salesforce and Zendesk are building deeper proprietary ecosystems. When you are the smaller player, lock-in works against you: open standards turn what happens if we want to leave into a non-issue. What I would dig into: open standards are commercially double-edged. They reduce switching costs for buyers leaving you, but also for buyers leaving competitors to join you, and the math only works if the second effect is bigger. The unresolved question is how Freshworks prices the MCP layer without contradicting the open narrative.
Bet 2: Outcome-based metrics (xLAs)
Freshworks redefined measurement from compliance-based SLAs to experience-based xLAs, a buyer-friendly metric tied to outcomes their customers’ bosses care about. What I would dig into: there is a measurement-pricing coherence gap. Freddy AI Agent is priced at roughly $49 per 100 sessions, an input metric, while the framework asks customers to evaluate on outcomes. The longer that gap persists, the more it undermines the xLA story. If I were on the team, I would move pricing toward per-resolution to align the commercial model with the measurement one.
Bet 3: Tier rollout reveals the go-to-market
Freshworks gated the Agent Studio on Growth and Pro plans, not just Enterprise, trading Enterprise-upgrade revenue for broader mid-market adoption. For their actual market position, that is the right call. What I would dig into: the inconsistency. Agent Studio is Growth-plus for Freshservice, but the Email AI Agent is Pro-plus for Freshdesk. Same parent brand, same launch, different tier-gating, which quietly teaches buyers two contradictory mental models of the company.
What I would want to know before any of this is real
I would be wrong to propose changes without saying what data would validate or invalidate them. Three questions: the MCP customer-adoption curve in the first six months, how many customers actually run xLAs as their primary operating metric versus a nice-to-have, and the Growth-to-Pro conversion rate within 90 days of activating Agent Studio. The numbers are not public, but they are the ones that decide whether each bet landed.