Moving Fast

Observations in the age of AI

With the proliferation of LLMs, the largest unlock for me has undisputedly been the ability to move fast.

Previously, moving fast seemed to be a moat, only achievable by people with some level of deep knowledge and skill in a particular area. The excuse “I have no context” was completely valid.

Now, seemingly anyone can move fast.

Just the other week I vibe coded payment features at xAI in Rust, a programming language I have no knowledge in and is notoriously complicated.

What would’ve taken a month to learn a new language, get context, and test took just a couple days.

While this is very empowering as an individual, I noticed the ability to move fast exposed weaknesses in the teams I work with.

Here are some of those observations on moving fast.

Sense of Urgency

While AI has provided the ability for people to move fast, it has not provided with the feeling they must move fast.

When given leeway, I noticed people will still dilly-dally and take their sweet time with things, even if they don’t need to.

Just because a person can move fast does not mean they will actually move fast.

I believe this is the difference of having an innate sense of urgency.

Personally, I have this vague sense of falling behind, which manifests as an obligation to be aggressively productive. Moments of doing nothing feels uncanny even if i truly need it. As such, I will happily move faster given the opportunity to do so.

Now, it’s totally fine for people not to have this feeling of what Oliver Burkeman coins as “Productivity Debt.” And it’s probably healthier to not be ruled by a productivity tyrant anyhow.

Regardless, having expectations that people will move fast because they can is not realistic. Not everyone wants to.

Agency

A big part of moving fast relies on agency - the less reliance on other people the better.

In a similar line to urgency, just because people have a tool to get high agency does not mean they actually will have higher agency.

Even though the barriers of high agency are lower than ever before, people with low agency will still shift the responsibility of their progress onto other people.

Recently, my team recently welcomed a new joiner and paired me to work on a new project with him.

What ensued was firestorm of slack questions from him for context that 1) I don’t have (because I am new with the project space) and 2) is easily searchable. And when I didn’t answer, he just sat there did nothing.

As someone who values moving fast, nothing sucks more than to be bogged down by actually stupid questions from someone who can answer them himself with a simple LLM chat.

The foolishness of having misplaced low agency is more exaggerated than ever before.

Blockers

There will always be blockers, in which progression solely relies on someone else.

For individuals in the age of AI who already have high urgency and agency, blockers become more apparent and frustrating. Accumulating increasingly large opportunity cost, blockers need to be removed immediately to maintain morale and momentum.

A reoccurring example I experience is getting code approvals.

In order to merge code and release it to production, I typically need someone to review my code and approve that it’s ok.

Now that I move at a higher velocity, code approvals become more frequently the main bottleneck.

If I don’t get approvals, I get stuck and can’t make anymore progress, becoming increasingly frustrated as I remind people on Slack to review my code every hour.

Just because a manager says “we move fast” does not mean the team actually moves fast. Actions mean more than words.

To me, being slow to unblock blockers eats away at a team’s legitimacy and claims of moving fast.

Incentives

Now that moving fast is easier, the need to align incentives become even more apparent.

Incentives are everything. If you want people to move fast, the best way is to figure out an incentive structure that rewards them when they do so.

In the example I gave previously on code reviews, the reason why it’s difficult for me to get code reviews is because of the incentives of others.

Given that there’s more pressure to move fast in one’s given project with AI, people are more focused on their own code instead of giving someone else code reviews.

There’s a flex for all the project someone’s built, not in how many they’ve reviewed.

As a result, my project gets frequent blockers and moves slower.

If this is multiplied across an org, it’s quickly apparent that overall progress begins to be hampered by conflicting incentives.

Aligned incentives are even more important now when people are moving fast and the costs of not doing are ever the more apparent.