Leading Without Authority
How senior engineers lead large efforts without org power: through credibility, coalition, conflict handled well, and framing every idea around the other person's goals.
For a long stretch of my career, no one has reported to me, and in that same stretch I have helped ship bigger things than when people did. That sounds backwards until you have lived it. The most senior people I admire have no formal power over the teams they move. They lead anyway. Early on I assumed that kind of influence was a personality you were born with. I have come to believe it is a craft, and like any craft you can get better at it on purpose.
TL;DR: Authority-free leadership runs on a handful of habits. Credibility you earn by being useful and right early. Coalitions you build before the decision, not during it. Conflict you treat as information instead of a threat. The willingness to own a hard call and carry its outcome. And the habit of framing your idea around the other person's goals, so that helping you is also helping them.
Credibility is the only currency you actually have
Without a title to lean on, the first thing I have to earn on any new team is the right to be heard. The only way I know to earn it is to be useful and right early, before anyone owes me their attention. Solve a small problem that was quietly annoying everyone, the flaky deploy step, the eval everyone ignored because it was too noisy to trust. Read the design document the room is arguing about and be the one who actually found the failure mode hiding in it. Bring a small working prototype of the agent loop to a debate that has been going in circles on opinion. Make a careful call about where the guardrail belongs that turns out well, and let the results speak.
It is slow, and it cannot be skipped. People do not adopt your idea because your reasoning is elegant. They adopt it because you have been right before and you were helpful when it cost you something. Credibility is a balance you pay into quietly for weeks and then get to draw on when it matters. I have never found a shortcut, and I have stopped looking.

Build the coalition before the meeting, not during it
The decisions I have lost were usually lost because I walked into the room and tried to win the argument live, in front of everyone, cold. The decisions I have won were mostly decided before the meeting started. I had talked to the people who would care, understood their worries, folded their good points into my idea, and walked in with allies who already nodded along.
This is not politics in the cynical sense. It is respect for the fact that people need time to think and do not like being surprised into agreeing. By the time the formal decision happens, it should feel like a confirmation, not a debate. A good idea with no coalition behind it loses to a worse idea that everyone already agreed to in the hallway. I learned that the hard way more than once.

Conflict is information, not a threat
I used to treat disagreement as something to survive. Now I treat it as data. When a peer pushes back hard on a design I am sure about, my first assumption is that they can see a constraint I cannot: a data freshness requirement the retrieval quietly depends on, an on-call burden, the cost of a model-routing choice that lands on their budget and not mine. Sometimes I am right and they are missing context, and the fix is to share it. Often they are right and I was about to walk into a wall I did not know was there.
The trick that keeps conflict useful is to argue with something concrete instead of opinions. A prototype, a latency measurement on the slow path, a small eval run over the two prompts, a cost comparison of routing to a self-hosted model versus an external one. Opinions versus opinions turns into a contest of confidence and seniority. Evidence versus evidence turns into a shared search for the right answer, and it lets the more junior person be right without it being a status fight. The goal is not to win the disagreement. It is to end up with the better decision and a peer who still wants to work with me.
Own the hard call
Influence is not only gentle. Part of leading without authority is being willing to put your name on an unpopular or expensive decision and carry the outcome either way. I have made calls that cost real time or money, where the safe move was to stay quiet and let the default happen. Taking ownership of those, including the ones that went badly, is a big part of how trust gets built. People extend influence to those who they have seen stand behind a decision when it would have been easier to hide.
Frame it around their win, not yours
The single most useful habit I have is also the simplest. When I want someone to adopt my idea, I describe it in terms of what they are trying to achieve, not what I am. The person who owns the roadmap cares about their outcomes. The person who owns delivery cares about hitting the date without surprises. If my idea helps them hit those, I say so plainly, and adoption stops being a fight.
It only works if it is true. You cannot fake alignment for long. But most good technical ideas genuinely do help the people around you, and we just forget to say it from their side of the table. When people can see that your design makes them successful, you do not have to push. They pull.
The takeaway
Leadership, I have decided, is not what you are allowed to do. It is what people let you do. No org chart grants that. You earn it by being credible, by bringing people along before the decision, by treating conflict as a way to find the truth, by owning the hard calls, and by making the people around you win. I have less formal power now than I did earlier in my career, and far more ability to make things happen. Those two facts are not a contradiction. They are the whole point.