2025 Recap and what's ahead in 2026

The year of agents is (almost) over! The hype going in was loud, but the end-of-year balance sheet is a lot more nuanced.

What we saw in 2025

Technology: Advancements like RLVR especially in coding and math, improvements in inference time and tool usage & harnesses drove the capability frontier.

Applications: In domains like coding and content creation, a host of apps from Gemini NanoBanana to Claude Code, demonstrably supercharged individual tasks and workflows in 2025.

Market: Capital allocation has increased and intensified, while early M&A shows how the market can sustain a lot more than the big labs and incumbents.

At the same time, real gains in bottom-line or top-line, outside AI-native hypergrowth startups - haven’t manifested for most organizations. This is not exactly unexpected. For experts who have studied general purpose technologies like computers and electricity, this is how change happens. To create trillions of dollars worth of real economic impact from AI, it will take closer to a decade, not a year.

What we’re looking forward to in 2026

RoI or Bust: Most Agentic AI pilots stall because of lack of RoI. At Poexis, we are obsessed with delivering 500%+ RoI in 90 days. For use cases like Account-based GTM and Agentic Demand Gen, that is what our customers have attained in 2025 - and it will only accelerate further.

Operational Scalability: For production GTM workloads that deliver topline and bottomline impact, point-to-point agent interactions simply fall apart. We spent a lot of time in 2025 building a new agentic operating infrastructure to get LLMs and the agents to do their magic, reliably.

Fragmentation: Early majority of the market has several point agents and solutions being piloted - if they have to move safely into production, let alone create clear RoI, they need to come together and work end to end, especially for GTM. That’s the focus for the Agentic OS.

Ambient Agents: Slowly but surely, early adopters are moving beyond turn-based conversations for Agentic UX into truly ambient agents on auto-pilot. We are long on frictionless ambient agents that integrate deeply into multi-user workflows.

Signal Engineering: With RLVR evolving into more domains, it is becoming clearer that companies who want to win in the AI age, need to invest in systems that capture and continually learn from their proprietary data and signals.

Building agentic systems that deliver GTM outcomes at scale - not one-off agents or apps - is a hard hill to climb. But it’s also exactly where we want to be, in lockstep with our customers.

Incredibly grateful to our customers and backers who are on this journey with us - let’s ring in 2026 and keep climbing! 🚀