In partnership with
Artificial intelligence is entering a different phase. The earlier generation mainly generated text, images, audio and code. Newer systems increasingly reason through problems, use software tools, analyze their own outputs and perform sequences of actions with less human intervention.
The International AI Safety Report 2026 describes major gains in mathematics, coding and autonomous operation. Leading general-purpose AI systems can pass some professional examinations, solve graduate-level science problems and complete increasingly difficult programming tasks. Yet their abilities remain uneven. They can perform difficult intellectual tasks while still failing on simpler multi-step problems. Hallucination, reliability and long-horizon planning remain significant limitations.
AI can build faster. Can your team decide better?
AI can draft the PRD and prototype the idea. Jira Product Discovery helps teams decide whether it belongs on the roadmap. Bring feedback and ideas together, prioritize as a team, and keep your roadmap connected to delivery in Jira.
Understanding AGI
Artificial General Intelligence, usually shortened to AGI, remains a debated concept rather than a settled engineering standard.
There is no universally accepted definition. A common interpretation describes AGI as a system capable of matching or exceeding human cognitive performance across a broad range of tasks. Other definitions place greater weight on learning new skills and solving unfamiliar problems without task-specific training.
This distinction matters. A model can perform extremely well across thousands of tasks without possessing the adaptability associated with human general intelligence.
Google DeepMind researchers have proposed measuring progress through performance, generality and autonomy instead of treating AGI as a single threshold. Under this framework, AI development can be studied as several levels of increasingly general capability.
Why AI capability is rising so quickly
Several forces are working together.
Compute remains one of them. Epoch AI estimates that training compute used by frontier language models has grown roughly fivefold per year since 2020. Its 2026 analysis also estimates that the available stock of AI computing power has been growing rapidly.
But larger training runs are no longer the entire story.
The International AI Safety Report notes that an increasing share of capability gains comes after initial model training. Reasoning techniques, reinforcement learning, tool use and additional computation during inference can improve performance without simply making the base model larger.
Hire Ava, the AI BDR built for enterprise
Ava is the first AI BDR to run outbound end to end, and you decide whether she runs autonomously or on copilot.
She finds leads or ingests accounts from your CRM, enriches them, sends personalized emails on behalf of your reps, follows up, handles replies and books meetings. Website visitor de-anonymization, intent signals, and a parallel dialer come built in.
A small team can manage her centrally for thousands of reps who never log in. Everything syncs two-way with Salesforce and HubSpot.
Ava runs outbound for companies like DoorDash and Grammarly, and one customer deploys her across 1,000+ reps. Ava is SOC 2 Type II audited, SSO and GDPR ready. Ava is how revenue teams grow pipeline without growing headcount.
Benchmarks are also becoming harder to interpret. Stanford's 2026 AI Index reports that frontier systems gained around 30 percentage points within a year on Humanity's Last Exam. At the same time, researchers have found errors and methodological weaknesses in several established evaluations.
The next stage may depend on agents
The shift from models to agents could matter as much as improvements in raw intelligence.
A conventional model waits for a prompt and produces an answer. An agent can potentially break a goal into steps, search information, operate software, write and test code, inspect results and continue working toward an objective.
Longer autonomous task duration therefore becomes an important measure.
The 2026 International AI Safety Report found that leading coding agents can reliably complete some tasks taking human programmers about half an hour. A year earlier, comparable reliable task horizons were below ten minutes.
Future systems could extend this from minutes toward hours, days and eventually longer projects. That development would affect software engineering, scientific research, finance, education, administration and many knowledge-based occupations.
Beyond language models
AGI research is also moving beyond text.
Future systems may combine language, vision, audio, video, spatial understanding, memory and interaction with physical environments. Recent research has examined visual general intelligence as another possible route toward broader intelligence. The aim is to understand environments through images, video and geometry rather than depending mainly on language.
Robotics remains much harder. Current AI can reason about many physical tasks without reliably performing them in the real world. Manipulation, physical uncertainty and continuously changing environments expose weaknesses that text benchmarks
How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads
The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.
