[Update Post] Revise ACE platform details and future roadmap

Author: Alex-q-z

_posts/2026-01-29-iclr-acceptance.md _posts/2026-01-30-iclr-acceptance.md_posts/2026-01-29-iclr-acceptance.md renamed to _posts/2026-01-30-iclr-acceptance.md

@@ -16,8 +16,6 @@ We release the code on [Github](https://github.com/ace-agent/ace) with easy repr
 
 **[[💻 Source code]](https://github.com/ace-agent/ace)    [[📚 Paper]](https://arxiv.org/pdf/2510.04618)**
 
-<br>
-
 ## What is ACE?
 
 Large language model agents are increasingly deployed in complex, multi-step environments: coding assistants, research copilots, data analysis agents, and more. But today’s systems are often *stateless*: each run starts fresh, discarding what was learned from previous attempts.
@@ -26,36 +24,37 @@ Large language model agents are increasingly deployed in complex, multi-step env
 
 > Instead of changing weights, ACE evolves the **context** supplied to an agent (its memories, plans, summaries, tools, and distilled experience) so the agent improves as it operates.
 
-<div align="center">
-<img src="/assets/img/ace-system.png" alt="Icon" style="width:700px; vertical-align:middle;">
-</div>
-
 Concretely, ACE provides:
 
 - **Empirical analysis and new understanding** of why prior context-evolution methods can fail silently or suddenly in multi-turn, long-horizon agentic tasks
 - **A unified framework** for capturing and structuring experience from prior executions without suffering from brevity bias and context collapse
 - **Evaluation across diverse agentic and reasoning tasks**, showing consistent performance and efficiency gains without retraining the underlying model
 
+<div align="center">
+<img src="/assets/img/ace-system.png" alt="Icon" style="width:700px; vertical-align:middle;">
+</div>
+
 Across coding, tool-use, and domain-specific reasoning benchmarks, ACE demonstrates that carefully engineered and continuously evolving contexts can unlock strong improvements in task success rate, efficiency, and robustness, thus pushing toward **self-improving AI systems** that learn from experience in deployment.
 
 If you are interested in how test-time memory, learning from experience, reflection, and planning interact with modern LLM systems, we think ACE will resonate with you! 
 
-<br>
-
-## ACE as a Context Engineering Research Platform
+## ACE as a Context / Self-Evolving Agent Research Platform
 
 Since releasing ACE, we’ve been thrilled to see growing interest from the research and open-source communities:
 
 - ⭐️ Active development at [github.com/ace-agent/ace](https://github.com/ace-agent/ace)
 - 📣 Talks and presentations at Hippocratic AI, HKU NAISS Lab, and more to come
 - 🤝 New research collaborations building on ACE as a platform for studying context-based methods and self-evolving agents at Stanford, MIT, ...
 
-Qizheng put a support graph here, include datasets already included, dataset being constructed, (same for frameworks like RLM, blabla)
-
-ICLR 2026 is just the beginning --- we see ACE as part of a broader research agenda around the new context learning paradigm. If you’re working on related problems or want to build on the ACE platform, we’d love to hear from you! 
+Over the next few weeks and months, we plan to release several major additions to the ACE platform across datasets, system features, and new research directions:
+| Datasets | Features | New Research |
+|-------|----------|------|
+| DDXPlus (medical reasoning)   | ACE + Agent Skills  | ACE + Meta-Memory Evolution |
+| BIRD-SQL (text-to-SQL)  | ACE + Recursive Language Models    | ACE + Tool-Augmented Context |
+| Terminal-Bench (terminal agents)  |   |  |
 
-<br>
+ICLR 2026 is just the beginning. We see ACE as part of a broader research agenda around a new paradigm for context-driven learning and test-time adaptation in AI systems. If you’re working on related problems or interested in building on the ACE platform, we’d love to hear from you! 
 
 ## Contact Us
 
-Shoot us an email at [qizhengz@stanford.edu](mailto:qizhengz@stanford.edu).
+Shoot us an email at [qizhengz@stanford.edu](mailto:qizhengz@stanford.edu).