# AI agent for development **URL:** \ **Category:** For Developers\ **Created:** [October 25, 2026, 10:59am UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846 "2026-10-25T10:59:36Z")\ **Posts on this page:** 7\ **Page:** 1
**Author:** ![gregd72002](https://forum.openwrt.org/user_avatar/forum.openwrt.org/gregd72002/32/20103_2.png) [@gregd72002](https://forum.openwrt.org/u/gregd72002)\ **Post date:** [October 25, 2026, 10:59am UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/1 "2026-10-25T10:59:36Z")
Hi, I have been increasingly often using AI to help me solve issues and port openwrt to new hardware. This works so well that now I am considering spinning an autonomous agent (Claude Code), give it required access and ask it to crack the bootloader so openwrt can be ported to currently unsupported routers. Anyone tried something like that? What I got so far is a new-old router (Draytek vigor 2862), wired a power button and tty to my linux box. I can control power, upload images through tftp, interact with it through tty - all from the box without touching the router. Next on my plan is to create some scripts to make it easier for the AI agent to control the power and boot process on the router. The idea is to get AI to get uboot to work first. Later to get it to build the DTS with as little input from me as possible. ---
**Author:** ![gregd72002](https://forum.openwrt.org/user_avatar/forum.openwrt.org/gregd72002/32/20103_2.png) [@gregd72002](https://forum.openwrt.org/u/gregd72002)\ **Post date:** [October 25, 2026, 3:25pm UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/2 "2026-10-25T15:25:42Z")
This works. I am blown away! My initial setup: - Ubuntu 24.04 - FTDI 3.3v ttl serial - HW-667 USB relay module for controlling power ![IMG_5497 copy](https://forum.openwrt.org/uploads/default/original/3X/b/d/bd459e485d1207f2ea8dbdcd8a8b23c38c0547c5.jpeg) Claude instructions: ```plaintext # Router investigation ## Goal Get Uboot-lantiq to run on the router from TFTP boot. ## Environment - Ubuntu host (192.168.1.10) - Router control scripts: /home/home/src/local_scripts/reboot.py /home/home/src/local_scripts/enter_tftp.py - Router TTY terminal is at /tmp/serial-console. It operates at 115200 speed rate. - Router default ip: 192.168.1.1 - Working directory: ~/router-investigation - Toolchain directory: ~/src/openwrt ## Safety - Power cycling is permitted - Flashing is permitted ## Workflow - Read next-steps.md at the start of each new session. - Read findings.md for established facts. - Save raw logs under logs/ and refer to them by filename. - Distinguish observed facts, hypotheses, and unverified claims. - Don't repeat completed tests unless needed to resolve a specific uncertainty. - Review existing openwrt patch that adds support for DrayBoot Vigor 2860 in here: /home/home/src/patches/26084.patch; Vigor 2862 might be similar - If changes are needed to uboot-lantiq package, create a new target; we might want to create a patch file later for the new target ``` After around 2h, around 50 trials, and ~50k tokens I got uboot on the board with no interaction from my end. The AI agent went through: - toolchain configuration & compilation - router discovery - boot process discovery and mapping of registers - compilation and patching of uboot to make it boot correctly on the device ```plaintext TFTP detected. RX: Filename: 'v2862_0.bin'\r RX: Loading:\r RX: \rD\rr\ra\ry\rT\re\rk\r \rD\rr\ra\ry\rT\re\rk\r \rD\r RX: done\r RX: Transferred = 524548 (0x80104) Bytes\r RX: Got .bin file\r RX: >> CodeStart=0x80024000\r RX: Go~~\r RX: \r RX: \r RX: U-Boot 2013.10-openwrt4 (Oct 25 2026 - 14:46:18) Vigor2862\r RX: \r RX: Board: DrayTek Vigor 2862\r RX: Unknown partnum: 11\r RX: SoC: Lantiq v1.1\r RX: CPU: 600 MHz\r RX: IO: 300 MHz\r RX: BUS: 300 MHz\r RX: BOOT: unknown\r RX: DRAM: 128 MiB\r RX: Using default environment\r RX: \r RX: In: serial\r RX: Out: serial\r RX: Err: serial\r RX: Vigor2862 # ``` ![IMG_5499](https://forum.openwrt.org/uploads/default/original/3X/1/6/160aeecdd0f0d658aeaf1000d88f17b6122e4711.jpeg) ![IMG_5500](https://forum.openwrt.org/uploads/default/original/3X/b/9/b9d57f703b12ff7fc10af3871cb26a826aab9e10.jpeg) ---
**Author:** ![olek210](https://forum.openwrt.org/user_avatar/forum.openwrt.org/olek210/32/10152_2.png) [@olek210](https://forum.openwrt.org/u/olek210)\ **Post date:** [October 25, 2026, 4:45pm UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/3 "2026-10-25T16:45:21Z")
I wouldn't be surprised if the AI Agent was learning from your latest posts on the forum and GitHub PR 😃 ---
**Author:** ![plappermaul](https://forum.openwrt.org/user_avatar/forum.openwrt.org/plappermaul/32/54365_2.png) [@plappermaul](https://forum.openwrt.org/u/plappermaul)\ **Post date:** [October 25, 2026, 9:09pm UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/4 "2026-10-25T21:09:48Z")
Agentic development for the OpenWrt Realtek switch target is running full auto since weeks. [https://github.com/openwrt/openwrt/pulls?q=is%3Apr+state%3Aclosed+author%3Al0rdg3x](https://github.com/openwrt/openwrt/pulls?q=is%3Apr+state%3Aclosed+author%3Al0rdg3x) ---
**Author:** ![rmandrad](https://forum.openwrt.org/letter_avatar/rmandrad/32/5_5575768a8748004e209b776fc1b2916d.png) [@rmandrad](https://forum.openwrt.org/u/rmandrad)\ **Post date:** [October 26, 2026, 8:45am UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/5 "2026-10-26T08:45:04Z")
this is great ... but i haven't see any leveraging any capability to ensure the code is secure both from a blue/red perspective. imho, this should be the priority bearing in mind what is going on embedding in pull requests security reviews is critical nowadays. ---
**Author:** ![gregd72002](https://forum.openwrt.org/user_avatar/forum.openwrt.org/gregd72002/32/20103_2.png) [@gregd72002](https://forum.openwrt.org/u/gregd72002)\ **Post date:** [October 26, 2026, 11:27am UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/6 "2026-10-26T11:27:32Z")
yeah, I can see the AI assisted development create new problems but resolves quite a few problems at the same time. It takes away some of the pleasure through... There was hardly a better feeling than spending weeks on a problem and get to the bottom of it yourself. Now the AI can do it in minutes... The blue/red perspective is the boring stuff for many, but certainly needed. Are there any guidelines on the approach? ---
**Author:** ![vochong](https://forum.openwrt.org/user_avatar/forum.openwrt.org/vochong/32/20001_2.png) [@vochong](https://forum.openwrt.org/u/vochong)\ **Post date:** [October 26, 2026, 12:03pm UTC](https://forum.openwrt.org/t/ai-agent-for-development/254846/7 "2026-10-26T12:03:16Z")
We could easily create a dedicated skill for the agent to run security audits from both blue and red team perspectives. Security auditing is already one of the strongest capabilities of modern AI agents, so integrating it into the PR workflow should be a natural next step.