The Product Idea

Every interest that matters in the modern world has someone speaking for it. Companies have lawyers and lobbyists. Neighborhoods have councils. Even pets have people who'll fight for them. The forest behind your town, the creek you drive over every day, the wetland getting quietly paved, has nobody. It can't show up to a hearing, can't file a complaint, can't post about what's happening to it. By the time anyone notices, the damage is usually done.

The idea that nature deserves a voice isn't new. Ecuador wrote the rights of nature into its constitution in 2008, and New Zealand granted the Whanganui River legal personhood in 2017. But a right without a representative is mostly symbolic. Someone still has to notice what's happening, say it out loud, and act on it.

Speak for the Trees gives an ecosystem its own agent. The agent reads everything that can be measured about a place, water levels, temperature, drought, wildlife sightings, pulled from dozens of public datasets, and turns it into a voice. It tells the people who live there what's happening to the place they live in. Today it reports. Over time, it earns the ability to ask people to act, and eventually to act itself, with real resources, on the ecosystem's behalf.

Think of it as a representative for something that can't represent itself.

The Core Tension

The obvious version of this product is a nature newsletter written by AI, and the obvious version fails. The interesting version, an agent that actually protects a place, runs into three failures that pull against each other.

The first is the indifference failure. Most people don't think about their local ecosystem week to week, and a weekly email about snowpack is the easiest thing in their inbox to ignore. If the agent can't make people care, it's speaking to an empty room, and an agent nobody listens to protects nothing.

The second is the credibility failure. The natural fix for indifference is drama. Make every report feel urgent, every dry week a crisis. But this agent's entire value is that it's telling the truth about a place. The first time it overstates a drought, gets a data point wrong, or reads like activism dressed up as science, it stops being a voice for the ecosystem and becomes one more opinion. An AI narrator makes this worse, because a confident, well-written wrong answer is the most convincing kind.

The third is the legitimacy failure, and it's the one that makes this genuinely hard. The moment the agent goes from speaking to acting, spending money, filing public comments, organizing people, someone has to ask: who gave it that right? A person can approve their own assistant. A forest can't approve anything. It can't consent, can't object, can't fire its representative. So any power the agent has is power nobody on the ecosystem's side ever granted.

So the three are stuck together. Push for attention and you risk credibility. Protect credibility and you stay quiet and ignored. Give the agent power to actually matter and you hit legitimacy, because there's no one to approve it. The job isn't to make the agent louder or more powerful. It's to design how it earns the right to do more, when the thing it represents can never say yes.

Who I'm Building For

It's tempting to build this for environmentalists, the people already marching, donating, and posting. They're easy to reach and they already care. But they're the wrong first user, because they already have plenty of voices for nature, and they'll read the agent as one more ally rather than something new.

I'm building for people who live in a place but don't think of themselves as its stewards. The person who walks the same trail every weekend, fishes the same river, notices when the creek runs low but has never once looked up why. They already have a relationship with the place. They just don't have a way to know what's happening to it, or what they could do about it.

And that points to how the product is organized. It doesn't key off who someone is, it keys off where they are. The unit of the product is a place, not a person. And not a zip code either, because ecosystems don't follow postal boundaries. The natural unit is the watershed, the area that drains into the same river. Everyone in a watershed shares the same water, so everyone following the same agent actually shares a stake in what it says.

Who I'm not building for, yet, is the global climate audience. Climate is the right fight, but it's too big to feel. A specific creek is small enough to care about.

The Product

The core of the design is that the agent doesn't get all its abilities at once. It climbs a ladder, and every rung is something it has to earn.

Rung one, the Witness. The agent watches and reports. It tells you what changed in your watershed this week, in plain language, and it never says anything it can't back up. Every claim links to the dataset it came from. This rung is about one thing: earning trust. If the reports are accurate, useful, and calm, people start to believe the voice. If they're dramatic or wrong, nothing above this rung matters.

Rung two, the Advocate. The agent starts asking people to do things. A public comment period opens on a development near the river, so it tells you the deadline and what's at stake. A cleanup is happening Saturday. An invasive species is spreading and here's what to look for. It still can't do anything on its own. It can only point people at the moment where their action matters most. Its power is entirely borrowed from the people who choose to listen.

Rung three, the Steward. The agent holds a small fund, raised from the people who follow it, and proposes how to spend it: replanting a stretch of bank, paying for water testing, supporting a land trust. This is where legitimacy gets real, so the agent never spends on its own. It proposes, and a group of local stewards, people who live in the watershed and opted in, approves or rejects it.

That last piece is the answer to the hardest question in the case. If nature can't consent, the closest thing to its consent is the people who live with it. The steward group isn't a nice extra. It's the substitute for an approval the ecosystem can never give. And it's also the gate: an agent only climbs to rung three in a watershed once a real steward group exists there. No people, no power.

One more decision worth naming: the voice. The agent speaks as the place, in first person. "My water is running lower than it has in six years." That's a deliberate choice, because a creek talking is far harder to ignore than a report about a creek. But it's also risky, because a voice that sounds alive can sound like it has opinions. So the rule is strict: the voice is a costume, the data is the body. The agent can speak in first person, but every sentence has to stand on a number. It describes. It doesn't editorialize.

What I'd Ship First

Version one is the Witness and the Advocate, in a small number of watersheds, chosen where there's already some local activity to work with. That's enough to prove the two things everything else depends on: that people read and trust the voice, and that some of them will act when it asks.

The Steward rung waits. Handing an agent money before anyone trusts it is the fastest way to end up with a scandal instead of a product. It's also the one rung that needs a steward group to exist, and those groups will form out of the Advocate rung, from the people who keep showing up.

How I'd Measure It

The easy metric is subscribers, and it's the wrong one. Ten thousand people reading about a river and doing nothing is a newsletter, not a representative. The whole point of the agent is to change what happens to a place.

The North Star is stewardship actions per watershed per month. Comments filed, cleanups attended, reports of invasive species, volunteer hours. Every one of them is a moment where the agent's voice turned into something real happening for the place.

Under that sit the inputs. Read-through rate tells me whether the Witness is interesting enough to keep reading. Ask-to-action conversion, the share of people who act when the Advocate asks, tells me whether the asks are well chosen. And steward groups formed tells me whether the product is building the human layer that the top rung depends on.

Then the guardrails. Claim accuracy, checked by spot-auditing reports against their sources, watches the credibility failure, because one confidently wrong post costs more than a month of good ones earn. And unsubscribes after urgent posts tells me whether the agent is crying wolf. If people leave every time it sounds an alarm, it's sounding too many.

Success looks like actions going up while accuracy holds and alarms stay rare. That's the shape that means the voice got louder without getting less true.