Perplexity launches its own take on AI agents for work

Perplexity launched always-on AI agents called Teammates, designed to fill multiple roles within a workspace and use existing projects, skills, and memory from its tools. The agents integrate with Perplexity Computer and are positioned for collaborative, ongoing large-scale work; the article provides no pricing, performance figures, or market reaction.

Analysis

The competitive implication is not the agent label itself, but a possible shift from selling model access to owning persistent workflow context. If users keep projects, permissions, and memory inside one workspace, switching costs could accrue to the platform layer and make model quality alone less differentiating. That is a conditional risk for Anthropic (ANT) and other model providers if enterprise buyers increasingly purchase bundled workflow outcomes rather than standalone model capacity; it could also pressure incumbent software vendors if agents displace paid seats or automate work those seats support. Conversely, agents that increase usage may expand demand for underlying models and cloud compute, so substitution and workload growth can coexist.

Near term, treat this as product-positioning news, not evidence of revenue displacement: there is no adoption, pricing, retention, security, or customer-outcome data here. Over the next 1–3 months, the important tests are paid deployment, repeat usage, enterprise controls, and whether agents reliably complete multi-step work. Over 6–18 months, persistent workflow ownership could influence software seat counts and model-provider bargaining power. The contrarian risk is assuming agents reduce labor or software spend quickly; integration, permissions, reliability, and user oversight may limit realized savings. No directional position is justified on this announcement alone.

Key Decisions for Investors

• No immediate trade in ANT: the announcement supplies no evidence of customer losses, pricing pressure, or a change in Anthropic’s financial outlook.
• Track enterprise adoption and renewal evidence over the next 1–3 months, especially paid usage, repeat engagement, security features, and customer-reported task completion; treat product claims as unverified until corroborated.
• Monitor enterprise software vendors for seat-count or pricing commentary alongside agent adoption. A durable decline in paid seats would strengthen a sector-underweight thesis; rising usage without seat contraction would weaken it.
• Revisit the model-provider thesis over 6–18 months if customers shift procurement from model access toward bundled workflow products. Falsifiers include sustained expansion in model usage, stable software seat counts, or evidence that agents require substantial human review.