The Lightchain roadmap for November 2026 to January 2027 is a three-month development plan that LCAI holders approved in a DAO vote. It focuses on two things: AI agents that live on the blockchain and can be called and paid per use, and a grants program for the independent operators who run the network's GPUs and nodes. The vote closed on November 17, 2026, with about 375 million LCAI for and none against.
This page explains what the plan covers month by month, why almost every step happens on the test network first, what the vote approved and what it did not, and how to check the progress on the public chain yourself.
The plan in plain words
Lightchain's development work runs in programs of three one-month sprints. Before a program starts, the development team publishes it as a proposal, and holders of native LCAI vote on it at dao.lightchain.ai. This is the second program. The first ran from August to October 2026 and put the core contracts in place: the ones that record AI jobs, hold the settings for each model and take payment per call.
The second program builds on those contracts in two directions:
AI agents. An agent is an AI service with its own name on the chain, such as atlas.lcai. Apps, and other AIs, can find it by that name, call it and pay it for each use.
Operator grants. Workers serve AI models on their own GPUs, and validators run the nodes that produce the chain's blocks. The grants program decides which of these independent operators receive a grant each month, under rules anyone can check.
November: a test network ready for real AI jobs
The targets in the next three sections are taken from the proposal. They are plans, not results.
Agents and grants both need a test network that carries real AI jobs, so the first month builds that.
Workers on the test network. At least two workers serving Llama 3 8B are staked on the test network, at least one on its own hardware rather than a shared server.
Every job checked. The disputer, the network's checking service, re-runs every job on the test network. On mainnet today it re-runs a sample.
The first agent. The agent contracts go live on the test network, and the first agent, atlas.lcai, is created in a single transaction. A public gateway lets apps and AI tools find an agent by its name.
Grants in rehearsal. Test operators go through a full practice month of grants. A third party reruns the numbers and has to reach the same result.
A safer network. After a test network outage on November 8, public endpoints are locked down and every service gets a backup connection.
December: agents you can call and pay
The second month makes agents usable, still on the test network.
Pay per call. Apps pay an agent in LCAI for each call, through the network's x402-inspired pay-per-call payments. Each payment is tied to a single request, so it cannot be used twice.
Agents that work with agents. One agent can hand a task to another.
Answers you can trace. Anyone can confirm that an answer came from the job it claims to come from.
A public demo. On December 12, an agent is called from Claude Code and paid by a script. The demo is recorded, and its numbers are published.
The first real grants run. The December run uses the fixed grant scripts, and its result is published.
In the same month, the agent contracts go to an outside reviewer. The test network also spreads its validators across seven or more servers at three or more providers, so losing two servers no longer stops the chain.
January: reviewed and ready for mainnet
The third month closes the review and rehearses the mainnet launch.
Review closed. Every finding from the outside review is fixed, and the report is published.
Every model checked. The disputer covers every model the network accepts. Once it does, penalties for wrong answers are switched on, on the test network first.
One worker build. An official worker build fixes the model files and settings, so every worker runs the same setup and their answers can be compared fairly.
Agents with names and mailboxes. Each test agent gets a page on the testnet explorer and its own mailbox, and the registration flow for .lcai names is tested with prices in LCAI and US dollars.
A mainnet dress rehearsal. The full mainnet launch of the agent contracts is rehearsed on the test network, and a go or no-go decision is recorded at the end of the month.
How a step reaches mainnet
Almost nothing in the plan goes straight to the main network. The test network works the same way as mainnet but runs on test coins with no value, so changes can be tried in public without risk to anyone's funds. Each mainnet step follows the same path:
The work is built on an internal development network.
It is deployed on the test network and rehearsed there within the sprint.
The testnet results are confirmed, with a transaction hash or address that anyone can look up.
The step gets its own vote or approval, where one is needed.
When the sprint is complete, its steps go to mainnet as one release, and the date is announced in the weekly update.
That is why the proposal sets no fixed mainnet dates. In its own words, rehearsing first means "mainnet votes are cast on something the community has already seen work."
What the vote approved, and what it did not
The vote ratified the three-month program, authorized the testnet transactions and test-operator funding it lists, and bound the development team to weekly updates. It did not approve the mainnet steps themselves. Each of these still needs its own vote or approval:
The handover of the protocol contracts. The contracts that hold the network's AI settings, jobs, workers and fee pool would move from the protocol multisig to the DAO's Timelock, as already happened for the treasury and LightDEX. Today a multisig that needs three of its five signers adds, prices and removes models.
The first change to model fees. Model fees would change in steps, under a schedule set out in a separate grants proposal that holders have not voted on yet.
Agents on mainnet. The agent contracts would be deployed on mainnet, then name claims would open and the first .lcai names would be registered.
The next core upgrade. A new version of the main job contract, written to fit a size limit the community agrees on in December.
The proposal adds one more limit. Any change that affects treasury funds, protocol parameters or mainnet contracts needs a separate governance vote, and no change to the plan may skip the testnet rehearsals or the votes.
How the grants are meant to stay fair
A grants program only works if people can see why each operator was paid. The plan builds the rules in public before any grant goes out:
Uptime that counts every hour. An hourly log records which workers are online, and uptime is measured against every hour a worker should have been online, not only the hours that were logged.
Tiers from real work. Grant tiers come from the jobs a worker actually served.
Hardware that can be checked. Workers report their GPU, so the hardware behind each grant can be verified.
Results anyone can rerun. The eligibility script, the weekly test prompts and every grants run are published. Each run ends in a payout root, a single fingerprint of the whole payout list, and anyone who reruns it should get the same one.
The rules themselves come in a separate grants proposal, which holders will vote on.
How to check the progress yourself
The plan is built to be checked rather than trusted. These are the places to look:
The weekly update. A written update every week on Discord and the governance forum, covering what is done, in progress and blocked, and what comes next.
The sprint reports. At the end of each month, what was delivered against what was planned, with reasons for anything carried over.
The testnet explorer. Every deliverable, rehearsal and payout run is referenced by a transaction hash or address on the testnet explorer.
The votes. This proposal, and every vote that follows from it, is public at dao.lightchain.ai.
What the plan promises, and what it does not
The proposal calls its dates "targets, not guarantees". The order of tasks and the technical approach may change during the program, and each change is reported in the next weekly update with its reason. If a deliverable is dropped, cut back or moved past January, the sprint report says so, with an explanation and a new target date.
Frequently asked questions
What is on the Lightchain AI roadmap?
From November 2026 to January 2027: AI agents built and tested on the test network, an operator grants program, and rehearsals of the related mainnet steps.
Did LCAI holders approve the plan?
Yes. The vote closed on November 17, 2026, with about 375 million LCAI for and none against. A passed proposal waits 48 hours in the Timelock before it can be executed.
When will .lcai names launch on mainnet?
The proposal gives no date. Name claims on mainnet come after the January rehearsal and need their own approval. The waitlist is open at agents.us.org.
Does the DAO control the models today?
No. Today a three-of-five multisig adds, prices and removes models, and every change is a public transaction. Moving those contracts to the DAO's Timelock needs its own vote.
Are workers penalized for wrong answers today?
No. Penalties, called slashing, are built into the protocol, but every rate has been 0 on both networks since September 2026. Under this plan they are switched on for the test network first, once every model is covered by checks.
Follow the plan
The full proposal is on dao.lightchain.ai. When the first weekly update names a testnet deliverable, look it up on the testnet explorer and see it for yourself.
This article is for informational purposes only and is not financial advice.


