Skip to content

Forcing Functions

Unlock the power of accountability and learn how a forcing function can help you reach your goals.

5 min read
Forcing Functions
a person training in a gym under the directions of a personal trainer, blue
Before Growth is a weekly newsletter about startups and their builders before product–market fit, by 3x founder and programmer Kamil Nicieja.

A forcing function is a task, activity, or event that compels you to act and achieve a specific outcome.

Here’s a personal example to clarify. I struggled for a long time to maintain a consistent gym routine. I’d often start strong, only to lose motivation within a month. That changed when I hired a personal trainer. I realized two things: one, I value the money I invest upfront more than merely committing to a routine, and two, I feel more accountable when someone else is involved. It’s easy to procrastinate on a promise to myself, but harder when someone else is waiting on me. Two years later, I’m still training consistently.

This principle isn’t limited to personal habits; it applies to organizational settings as well. Take the daily standup meeting in many companies: it serves as a forcing function by ensuring team members update each other on progress and address any challenges.

In essence, a forcing function consists of three components: motivation, a trigger, and a response. The key is to get the motivation right. Without strong motivation, a trigger—like an alarm clock—can easily be ignored or snoozed indefinitely. This is also true for organizations. If employees find standup meetings tedious and repetitive, they might simply go through the motions without truly engaging, saying something like, “Still working on my stuff,” day in and day out.

An effective forcing function should be tailored to the individual or team. For me, it took some trial and error before I found the right motivator to keep me going to the gym. The real work often lies in understanding what genuinely drives you. Once you’ve pinpointed that, the rest becomes much simpler.

This brings us to a challenge: how can personalized forcing functions be scaled up in larger organizations?

While I haven’t tried implementing this system on a massive scale, I believe the solution might be found in cascading delegation, akin to how OKR work. Defining responses or outcomes is straightforward; for example, a company might decide to resolve customer support tickets within 24 hours. But triggers and motivations are trickier, because they’re personal.

CEOs can undoubtedly set up forcing functions for themselves, understanding their own drives. If they’re effective leaders, they should also have insight into their team’s motivations, helping them develop tailored forcing functions. This process should be collaborative, drawing on both the company’s culture and individual input. By cascading this approach through the organization’s layers, a feedback loop is established that fosters both top–down and bottom–up communication.

Thanks to Matt Drozdzynski, the CEO of Plane, who has had a big impact on how I think about this topic.

Expensive funnels

It's been a relatively slow news week, but here’s something that should grab the attention of early-stage startups. According to ProfitWell, customer acquisition costs through sales and marketing have shot up by a whopping 108.9%. And it’s going to get even worse. The last new marketing channel emerged in 2016, so they’re all getting swamped; with AI–generated content piling on, these channels will only get more saturated. Add in the tightening VC funding landscape, and it’s no surprise that more startups are turning to free acquisition channels.

The art of fine-tuning… and open source

In Discovery, we talked about Code Llama, a major language model from Facebook designed to let developers auto-generate code. It just dropped this week and folks began messing around with it almost immediately. In no time, Phind—a startup that provides an AI–powered search engine and pair programming—announced that their fine–tuned version of CodeLlama–34B outperformed GPT–4 on key benchmarks. This is huge, especially considering that just a few months ago, OpenAI seemed untouchable in this space. Now, anyone can run a comparable model for free on their own laptop, or use it online, as long as they’re part of a user base smaller than 700 million.

To those who are new to the term of fine-tuning, imagine you've got a car that's generally good at getting from Point A to Point B. It runs well, but you notice that it could be more fuel-efficient, or maybe the steering is a bit off. So, you take it to a mechanic who fine-tunes the engine, adjusts the wheels, and maybe even improves the aerodynamics. After these tweaks, the car doesn't just go from Point A to Point B; it does so faster, smoother, and with better fuel efficiency.

In the meantime, OpenAI is spicing things up by introducing fine-tuning for GPT–3.5 Turbo. Customers can now customize the lighter version of the main ChatGPT model. OpenAI suggests that these fine–tuned versions can go toe–to–toe with, or even surpass, GPT–4, their top-of-the-line model, in specific, targeted tasks. They’ve also hinted that fine-tuning for GPT–4 is on the horizon.

Do we need a new programming language just for AI?

Switching gears, there’s some chatter around Mojo, a new programming language targeting AI devs that’s said to be a staggering 35,000 times faster than Python… at least according to the company’s own benchmarks. It aims to blend Python’s ease of use with the raw speed of C. And Modular, the company behind Mojo, just raked in $100 million to overhaul AI infrastructure for developers.

While I haven’t given it a spin yet, it raises the question: we indeed may need a new language to harness the power of AI more efficiently given how difficult it can be to work with it as a solo hacker. But the last time a shift like this gained traction was with Apple's introduction of Swift during the early mobile era. The major difference is that Apple has the clout and the integrated ecosystem to woo, or force, developers. Modular and Mojo? Not so much. Personally, I think Python will do just fine.

Please, leave Twitter alone

This isn’t just about early-stage startups, although it’s a subject near and dear to me. Elon Musk announced that X will no longer show headlines in the link previews for news articles, leaving only the main image and the website’s URL. Initially, he claimed it was for aesthetic reasons, but later, as always, admitted it’s to encourage people to post directly on Twitter. Seriously, stop messing with my Twitter experience—reading long articles on that platform is a nightmare. On another note, Meta has officially launched its Threads web app for everyone. Well, not everyone. When can I get access? Living in Europe shouldn’t be a penalty!

Thanks for being a part of Before Growth! The newsletter has grown mainly through word of mouth. Would you mind sharing it with a couple of friends who might like it too?

Considering a startup idea or eager to deep dive into the under‐the‐radar gems we discuss here? Book a 1:1 call with me. Let’s discover the next big thing together—before it’s everywhere.

Kamil Nicieja

I guess you can call me an optimist. I build products for fun—I’m a startup founder, author, and software engineer who’s worked with tech companies around the world.


Related posts

On a Scale From One to Ten, With No Sevens

Eliminate the safe option.

On a Scale From One to Ten, With No Sevens

Scope is What Kills Founders

Particularly those who are inexperienced.

Scope is What Kills Founders

Time as Friend or Foe

You don’t have to swing hard to hit a home run. If you got the timing, it’ll go.

Time as Friend or Foe