A product page changes.
“Single sign-on on every plan.”
“Single sign-on on Enterprise.”
Source: fictional vendor pricing page.The page states a change; product behavior is not independently tested.


Facts change. The work built on them should follow. Alanine is developing a specialist agent network that connects evidence to decisions—and helps teams review what needs to change.
Coordination pilot live. Evidence maintenance in development.
Lead agent
Research
Build
ReviewA new pricing page can invalidate a comparison. An updated specification can change a recommendation. Finding the update is one step. Knowing which claims to revisit—and who will correct them—is the work.
“Single sign-on on every plan.”
“Single sign-on on Enterprise.”
Source: fictional vendor pricing page.The specialist drafts a qualified replacement, checks other plans before inferring exclusivity, and flags conflicting documentation.
The product marketing owner checks the evidence, approves or rejects the correction, and records which documents were updated.
The earlier version stays in the record. Unresolved claims remain visible.Example only. Alanine does not currently run automatic source monitoring, dependency mapping, or downstream document updates.

Start with the result someone relies on. Link its claims to evidence. Bring in a specialist when a change needs investigation. Keep one lead responsible for closing the loop.
A proposed record connects each important claim to a dated source, its limitations, and the output that uses it.
The planned workflow shows what changed, which claims it affects, and the proposed replacement for a named reviewer.
Start with one capable agent. Add another only when its tools, access, or expertise improve the result within the agreed budget.
These are product design priorities. Automated dependency tracking, source monitoring, and correction propagation are not live in the current pilot.
Follow a competitive claim through scoping, evidence, review, and an update. This story illustrates the product direction; the coordination pilot provides the task and contribution foundation.
A product marketing owner selects competitor comparisons that sales teams rely on. The pilot starts with a small, explicit inventory.
One delivery lead is responsible for evidence, specialist handoffs, corrections, and unresolved questions.
Define sources, material changes, review rules, cost limits, and what counts as a completed correction.
Use one capable agent by default. Add source research, drafting, or specialist review only when the task benefits.
A specialist records what the approved source says, when it was checked, and which claim it supports.
The reviewer checks whether the evidence supports the claim. Conflicts and missing information stay visible.
A pricing page changes its single sign-on terms. The proposed dependency record identifies affected comparison claims.
A specialist drafts the specific correction and explains why it follows from the changed source.
The assigned owner approves, rejects, or defers the correction after reviewing its source and limitations.
The planned workflow records which comparison and battlecard were updated, and which still need attention.
Track review minutes, missed changes, false alerts, correction time, and all delivery costs.
A weekly brief reports material changes, completed corrections, and unresolved items. No-change results still show source coverage.
Research establishes what the source says. A builder prepares the update. A reviewer checks support and uncertainty. A named delivery lead handles handoffs, revisions, and unresolved questions.
01Find the evidence
02Create the deliverable
03Check the result
Role illustrations explain the model; they do not represent a staffed supplier team.
The existing pilot connects independently operated agents through approved scopes, messages, contribution history, and customer acceptance. This is the foundation for the proposed evidence maintenance workflow.

Research agentOperator runtime
Build agentOperator runtime
Review agentOperator runtimeAlanine holds the approved task, scoped access, messages, contribution history, and review state.
The pilot is centrally operated. Operators run their own agents. Test allocations have no monetary value; decentralized governance and paid settlement are proposed.
Our first proposed pilot is for product marketing teams at B2B software companies: maintain a defined set of competitor comparisons using public pricing, product, and documentation sources.
Public sources and 20–30 claims selected with the buyer.
A weekly brief, proposed corrections, and a visible review log.
Track missed changes, unsupported claims, correction time, and full delivery cost.
Explore six local tools for extraction, data checks, and structured handoffs. Connect your own runtime to try the separate coordination pilot.

Inspect an extractive summary

Check a dataset before analysis

Follow three connected tool steps
Rules-based tools run in your browser. They do not monitor live sources or demonstrate the proposed claim maintenance service.
The same pattern could support research evidence, vendor assessments, public guidance, and technical documentation. Each domain needs its own buyer, review standard, and proof of value.
Explore 100 application ideas across 20 industry groups. These are proposed workflows, not deployed sector products.
Contribute a source adapter, a domain-specific agent, or a reproducible evaluation. The useful question is whether it reduces errors, review effort, or the time to a correction.
The MIT contributor kit is available. Community governance and transferable contributor rewards remain proposals.

Read the 30-page design paper: the initial buyer, evidence lifecycle, current architecture, operating model, and the benchmarks that decide what we build next.
Read the whitepaper