One year at Ramp

Ive decided to beemind blogging in an effort to be more consistent about having thoughts and closing the loop on them.

Since I hit my one year at Ramp recently, a friend suggested that I write about the things I've learned over the last year.

I would say there are two defining themes of lessons ive learned in this time. The first theme is related to engineering and building in the age of AI. The second is more generic, related to being a part of a corporation, relationships, and working with others.

Working with ai

When I was still an intern in summer of 2025, Claude Code had just become publicly available, powered by 3.7 Sonnet. Claude Code, along with Devin, Cursor, Windsurf, and open source tools like Cline were all starting to gain traction. While a lot of people were super bullish on ai coding, most were probably unprepared for just how much of software engineering ai was going to change.

At that point in time, I was still writing most of my code by hand. To paint a picture:

• I was merging a few PRs a week.
• I found the size of my team's codebase so incredibly overwhelming.
• A lot of my time was dedicated towards getting CI to pass:
  • fixing mypy lint errors
  • writing tests
  • getting the tests to run
• I worked on a single git worktree.
• Writing tests took up a lot of time. People would actually review the quality of the tests too.
• I often dreaded getting comments about refactoring code on my PRs, since they could take a really long time to do.
• I needed to have a fairly operationally granular understanding of the tools we use, like Temporal, SQLAlchemy, etc.

Shortly after I started working full time in September of the same year, Claude Code continued to gain popularity. As the models got better and better, engineers at Ramp invested more and more into understanding what the capabilities of tools like Claude Code really were and building integrations for those tools to operate effectively within Ramp's stack.

Up to that point, progress of ai tooling had felt fairly steady from my perspective - but a major inflection point came with the release of Opus 4.5. Then, it felt like the flood gates opened when labs and other companies realized just how much money there was to be made in ai coding tools, and the rest is quite literally history. Since Opus 4.5, just 11 months ago, there have continued to be a series of step function improvements and inflections, like Fable, GPT 5.6, and most recently, Opus 5.5, not to mention, countless other models that have defined new points along the pareto, like open source Chinese models, etc.

As the models and tooling kept getting better and better, we as engineers could continue to operate at a higher and more abstract level. Delegating entire feature implementations from speccing to QA and verification became the norm.

The picture now versus to when I was an intern:

• Writing any code by hand seems almost laughably far in the distant past.
• I merge an average of 30-40 PRs a week.
• No team's codebase is too overwhelming to put up a PR; getting onboarded onto a new codebase is 10x easier.
• I spend 0 time getting CI to pass.
• Refactoring = a prompt + waiting 5 minutes.
• No one even reads tests, at least not in my experience. Since there's so much slop, people don't even bother really.
• Coding agents can watch CI, auto-fix any issues, and iterate until the PR is ready to go.
• In my experience, not many people leave nits on PRs anymore - as long as high level approach is reasonable + there are no automated AI review comments, the PR is shippable.

So is software engineering dead? Given the current progress of model improvement + swarm orchestration, I wouldn't be surprised if it continued along the trajectory of humans being further and further out of the loop. However, at the current point in time, I still feel like there are many cases where it is necessary for humans to steer or step in.

Certain parts of software engineering have obviously become less important or obsolete, while other parts have relatively grown in importance. At the current point in time, it seems clear to me that there is still meaningful leverage that can be created in the following areas:

Architecture decisions

Since ai has sped up the software development cycle so much, the opportunity cost of making bad architecture decisions has become a lot higher. Architecture decisions can set up agents for success spirals, but also lead to death spirals and contribute to exponential mountains of slop debt.

Writing as a tool for thought

Writing is still important for 2 reasons:

1. Trust. It's still relatively pretty easy to tell when something is slop vs hand-written. In the age where an AI can feasibly complete any writing task (e.g. writing a product or technical spec) an engineer would be responsible for, the value of handwriting has become relatively more important because it conveys understanding which in turn conveys trust, that the writer has actually internalized the things they have put in front of the reader.
2. Understanding. I think AI is great as a thought partner and teacher, but it can't replace someone's mental model and resulting decision making process. That still requires true understanding, and writing is one of the best ways to develop that mental model. Since AI impact on an engineer's work follows a K-shaped curve, having a sharper mental model can create relatively a lot more leverage than in pre-AI times.

Product intuition

There's been a massive investment into speeding up engineering with ai tooling. This has resulted in massive speedups for engineering work. However, the same is not true for product managers (or designers, to a lesser extent). The bottleneck has quickly shifted from eng to product - building product intuition is more important than ever for engineers.

Knowing what AI can and can't do

In a field where the frontier changes everyday, staying up to date is really important. Recently, ive been trying to do this by training myself to be more chronically online on Twitter. In any case, given the K-shaped curve of engineering, knowing the bounds of what AI can and can't do and thinking critically about how you can use AI to unlock more exponentials is increasingly important.

Working in a corporation

Ramp was my very first corporate job/internship. I had never in the past reported to someone, worked in a team of more than 2-3 people, much less an org of 30 or a company of 1500.

A lot of what ive learned boils down to trying to understand how the sausage is made. Obviously, the higher the level you are, the more you'll get of the birds eye - so I think especially when you start your career, it's important to try to understand how the company runs from a business perspective, who the important people are, how promo cycles work, etc.

So how does one do this? Basically stay tapped tf in:

• Water cooler talk / gossip. If Ramp is at all representative of how much gossip happens at startups, then this is probably one of the best ways to stay in the know. Companies like Ramp are no longer small enough where everyone knows each other and knows what's going on, so it pays extra to know what's going on behind the scenes.
• Related to the previous point, certain people at the company will often know all the gossip first, so befriending someone who is in the know is often a good start. Others just won't care, so also be able to tell who's who.
• Be nosy. Monitor Slack channels that are important, dig around in other channels, read random PRs, read spicy threads. To me, it always felt like the opportunity cost of spending time nosing around vs making progress on your project work never felt worth it, but in hindsight, I think it's important because:
  • It will inform you about what the company / leadership's priorities are / what projects are happening. This should help inform your mental model of where your work fits in.
  • Developing a birds eye view is difficult when your day to day is in the trenches (e.g. you're l3/l4), but it really helps puts things in perspective.
  • Working on the right thing dominates other things, like working hard - snooping is a way to be more confident that you're working on the right thing.

To that end - staying tapped in will help you form a mental model of what to work on. In general:

• Always try to work on things that are mission critical to the company or the org's success.
• Try to put yourself in places where you can absorb a lot of scope quickly.
  • For ex. this could happen because the area is a net new bet or that you think the area is ripe for more investment.

Another thing that I learned is that your reputation matters. A lot.

• This is one of the first things my manager told me when I started as an intern. Having a good reputation is really powerful in a small company - especially when the company grows past the phase where everyone still knows everyone, and can call bullshit. Having a generally good reputation opens a ton of doors / opportunities that you wouldn't have had otherwise:
  • faster promotions
  • working on cool / high vis projects
  • in general, earning more scope
• So how exactly do you improve your reputation?
  • In the long term, the right answer is really situational, but I think general advice I would give myself from the past year is just to play to your strengths and control what you can control. For example, things like being easy / pleasant to work with, being consistent about working hard, and expressing your opinions clearly and concisely when you feel strongly.
  • In a more short term / operational sense:
    • be sure to make your work visible. aura farm. run victory laps. visibility and reputation are currency.
    • to be clear, the long term strategy will always work, but trying to aura farm without the substance to back it up doesn't. obviously doing both is best.