Gespeichert in:
| Hauptverfasser: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2605.05980 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914539143954432 |
|---|---|
| author | Sui, Yuan Chen, Yulin Li, Yibo Jiang, Xue He, Yufei Dong, Yihong He, Xiaoxin Gao, Tianyu Hooi, Bryan |
| author_facet | Sui, Yuan Chen, Yulin Li, Yibo Jiang, Xue He, Yufei Dong, Yihong He, Xiaoxin Gao, Tianyu Hooi, Bryan |
| contents | When language model agents tackle complex software engineering tasks, they often degrade over long trajectories, which we define as *agent drift*. We focus on two recurring failure modes *overthinking* and *overacting*, i.e., where the agent repeatedly reasons over information it already has, and where it issues tool calls without integrating recent observations or acquiring new evidence. In this paper, we introduce TACT (Think-Act Calibration via activation Steering), to detect and mitigate agent drift in the residual stream before it surfaces as a behavioral failure. In specific, we label trajectory steps as overthinking, overacting, or calibrated, and find that their hidden states can separate linearly along two *drift axes*, pointing from calibrated behavior toward each failure mode (AUC $\approx$ 0.9). To mitigate agent drift, we project each step's activation onto these axes at test time and pull drifted ones back toward the calibrated region. Experiments show that TACT outperforms unsteered baselines across SWE-bench Verified, Terminal-Bench 2.0, and CLAW-Eval, lifting average resolve rate by $+5.8$ pp on Qwen3.5-27B and $+4.8$ pp on Gemma-4-26B-A4B-it while cutting steps-to-resolve by up to $26\%$. These gains frame agent drift as a steerable direction in the residual stream, and position TACT as a viable handle for reliable long-horizon agents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_05980 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | TACT: Mitigating Overthinking and Overacting in Coding Agents via Activation Steering Sui, Yuan Chen, Yulin Li, Yibo Jiang, Xue He, Yufei Dong, Yihong He, Xiaoxin Gao, Tianyu Hooi, Bryan Artificial Intelligence When language model agents tackle complex software engineering tasks, they often degrade over long trajectories, which we define as *agent drift*. We focus on two recurring failure modes *overthinking* and *overacting*, i.e., where the agent repeatedly reasons over information it already has, and where it issues tool calls without integrating recent observations or acquiring new evidence. In this paper, we introduce TACT (Think-Act Calibration via activation Steering), to detect and mitigate agent drift in the residual stream before it surfaces as a behavioral failure. In specific, we label trajectory steps as overthinking, overacting, or calibrated, and find that their hidden states can separate linearly along two *drift axes*, pointing from calibrated behavior toward each failure mode (AUC $\approx$ 0.9). To mitigate agent drift, we project each step's activation onto these axes at test time and pull drifted ones back toward the calibrated region. Experiments show that TACT outperforms unsteered baselines across SWE-bench Verified, Terminal-Bench 2.0, and CLAW-Eval, lifting average resolve rate by $+5.8$ pp on Qwen3.5-27B and $+4.8$ pp on Gemma-4-26B-A4B-it while cutting steps-to-resolve by up to $26\%$. These gains frame agent drift as a steerable direction in the residual stream, and position TACT as a viable handle for reliable long-horizon agents. |
| title | TACT: Mitigating Overthinking and Overacting in Coding Agents via Activation Steering |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2605.05980 |