building software that "helps" people.

I've spent over 10 years at this point working on apps that are for "social good", or at startups, or consulting for these startups. A very common remark is something like, "I want to do something that helps people", or "I want to build a tool that is actually helpful".

I've tried to build many of these things myself, and have looked long term at a lot of the software that aims to do this. This, in a deeply shortened ramble (that contains most of my career, hundreds of books, and dozens of software projects) that holds what I've come to know.

Social good, and help are two loosely defined terms, even on their own that vary wildly depending on where you sit in the political spectrum. Put simply, building software and saying it "helps" people is extremely easy and socially advantageous.

Building software that helps people concretely and measurably is a feat few have genuinely attempted, and fewer still have succeeded at.

For the purpose of this diatribe, note that I am referring to social good software that aims to relief or remediate human suffering in some way. This doesn't include enterprise software that makes spreadsheets easier or solves only a business problem, but theoretically does include software that is touted to help people indirectly (like Flock LPR cameras).

For ease, I'm narrowing it down to consumer facing software like Crisis Text Line, Headspace, Crisis / Safety apps, and others in that similar space. I will reference occasionally those other products for points of reference, but not as the focus of the lessons.

When we start by defining help, we need to start by defining how help is measured and who measures it. Is it actually beneficial to use engagement methods to get people to complete specific flows? Do the standard metrics that we use to determine the health of a piece of software encapsulate truly what it can mean to help someone? Often, what measures engagement in software is tied to revenue - not always end user benefit. Sometimes these can be conflated, but often not.

Let's complicate it further - who is qualified to evaluate that a user is helped? Think of an app designed specifically for mentally ill users. Two different profiles, a user with bipolar disorder in a manic episode, and someone with C-PTSD may have entirely different user feedback responses to a tool.

If a user in a manic episode really likes a tool and is in an elevated mood, they may rate it way too high. If a user with C-PTSD is triggered and otherwise in a mood of irritation, they may rate it too negatively And that is just capturing information within the moment.

Long term data capture about life outcomes is harder than people expect, and comes back through distorted biases of what the end user wants the surveyor to believe. However, it's necessary to actually center on if the intervention worked in the moment, or was able to induce long-term change.

This is a major issue with rehab centers, where people in recovery often lie on follow-up calls. Not from bad faith, but from shame about not staying sober. Otherwise, they don't respond at all.

It's also hard to know what each individual companies measure as helped: is it interaction or viewing an article? Is it engagement with that article, or a follow-up email that checks in on if someone was able to change an outcome with the information they learned? Defining the methodology of how we consider people helped is almost as important as defining what "help" is.

We can start with the example of most revenue based software: high usage is considered better, and often times product developers want people to use their tool as an ingrained habit or action. This often results in trying to maximize DAU (daily active users), or MAU (monthly active users). Some tools can also be paywalled from users, or placed behind security infrastructure that may make it harder for users to engage with.

A great example of this is mental health apps: they are incredibly widely accessible, but early research of them shows they may only work in a small-ish group of people. They also provide a relatively minor reduction in anxiety and depression, which is statistically significant but perhaps not actually long-term indicative of good habits instilled in the user through the software.

It can also be argued that most users are just participating in building "engagement" streaks, but not actually developing the underlying habits those engagement streaks are trying to promote.

This is where analytics and intervention development when it comes to software can get genuinely complex.

What does it mean to "help" a user? Going to the earlier examples - is engagement help? Is reading an article that may give them information, "help"? Is it prolonged usage of a tool or inversely negative engagement of a tool? There's also instances in which a party believes something is helpful to a person, and that person would not believe that intervention to be helpful (e.g: just think of the COVID vaccine rollout).

When figuring out what help means, there is already so many layers of nuance, autonomy, philosophy, intertwined with the necessity of an academic focus with none of the funding or resources to accomplish that full goal.

Is it any wonder then, that most software in this space seems to have a net-zero effect? Or if there is an effect, it's in minor isolation so much so that it rarely ever ripples into mass systemic change.

It's hard to know if the implementation of suicide hotlines has actually lowered the suicide rate, or if the prevalence of mental health apps has done anything to stem the mental health crisis that's been rising in the U.S among younger generations, despite wider awareness and accessibility of resources.

Figuring out if software genuinely helps people is a clearly unsolved problem, and users and their patterns can be unreliable narrators of their own lives and inner world.

These are not unsolvable problems though, and I believe that complexity is often what creates beautiful engineered solutions. The answer to a lot of this is already in the bones of product development methodology, in combination with program (not meaning software but methodology in the human service space) evaluation techniques.

It also means releasing a bit of our ideas on what makes a product helpful: separating usage metrics from frameworks that help measure depth and long-term impact instead of activity or interactions more meant for UI/UX refinement.

Determining the long-term impact is an entirely separate framework from the product, but also deeply required to make it anything close to effective.

This is before we get into what, "helped", truly means. Some of the most beautiful parts of software and its' ongoing development is that it forces us to look inwards.

Most of what we build for others is really a reflection of us, and our own values. Is helping a group determined primarily by that self-reporters from that group or alternate metrics? Is helping something that should be determined purely by the scale of the intervention or its' depth?

Many philosophical movements within software have always existed, and these factions feel more political than ever in 2026. While software has always had political connotations and implications, this year feels the most "corporate" in a way that software has never been.

Defense tech startups are now the hot places to work as opposed to areas where ethical consideration should be applied. Palantir and Anduril of 2026 is the Google in 2012, or Uber in 2018.

Would I consider Palantir or Flock to be building software that "helps" people? Uh, fuck no. But certainly, there are people with bigger pockets and more influence than me that do genuinely believe in these companies and their philosophy of help. I can be cynical about it, but I cannot discount it.

Tech CEOs are also more active than ever in political meddling, and the face of implications of technology is in the forefront of the public. Most people are talking and worried about AI taking their jobs, hearing that computer science is no longer an easily accessible middle class field. Gen-Z understands that tech CEO's are political figureheads more than they are naive & ambitious nerds who designed an algorithm and fell into "billionairism".

You can see these factions about what help looks like deepening today, and different philosophical movements within the startup & software industries that give their best answer to what "help" is.

For effective altruists, it was minimizing death per value investment. For defense companies, it is enabling surveillance and logistics towards the national interests of the United States. For generative AI companies, it is access to information that will enable more creativity or productivity from its' controller.

To all of these companies, helping is an end goal that they believe in as a philosophy of encoding business or technical values.

But, effectively and non-philosophically judging what help is, is incredibly hard. Help to each person is also so context dependent, and needs to be deeply cocooned to them while also honest, scientifically backed, diversified, ideally steeped in community, and is often a non-linear and frustrating process.

Help is more a marketing token than it has ever been a concrete goal, backed by business values and given the levity to make shareholder decisions. Help is an end-ideal in software companies they wish to sometimes achieve as a side effect of pushing their business values and philosophy, but it is not often a direct and main goal established with clinical zeal and rigorous testing the same way the non-profit human services sector may approach it.

Even when we may have a defined goal of help, measurements, and research to back it up: that doesn't always mean that software can get people there.

Anyone who has ever helped a person in real life knows that it is the opposite of a clear emotional panacea that many products need to be. Often, pushing people to do something that is helpful for them can be painful.

The desire to change and adapt to change needs to intrinsically come from within the user, enough to motivate them through the helpful flows without alienating them. This dance: pushing a user, the autonomy the user has, and the failure of the built world and business landscape, is often where we lose people when we need to hold onto them with everything we can.

Software products that want to be adopted at scale need to be stupidly easy.

They need to be culturally relevant to the person adopting them, and in clear benefit or value to the adopting party. To help, you also need trust and reputation.

In other words, the bar for software to genuinely help someone in a long-term beneficial way can often be …. at odds with product design and development requirements that VC-backed software companies provide. It also has to be deeply, deeply linked with strong incentives aligned with the motion of the world.

I have been the co-founder and CTO of a company that has builds software that estimates the resilience of a building to natural disaster for around 3.5 years.

The highly condensed thesis was, we can save lives and reduce displacement from climate change by retrofitting homes in high risk areas. We can give insurers more business and resilience by letting them figure out which of these homes are resilient.

Originally, we gave this information directly to homeowners. The homeowners, although somewhat interested, felt often too far away from the fear of a natural disaster and too untrustworthy of a startup or the financial benefits to do the mitigation work required.

Then, we started selling directly to insurers as a complimentary underwriting layer, meant to be used in conjunction with a NATCAT (natural catastrophe) model - one that determines the natural disaster risk of a property based on location and factors, but rarely the characteristics of the building itself.

Insurers wanted a guarantee that homeowners would do the work if provided the info, and homeowners wanted a guarantee that insurers would not fuck them in the first place.

In this way, although the software is designed to help both parties to align their incentives (not losing your home or insurance in a wildfire-prone area), we still struggle often to meet the reality of the value proposition in the current economy and with the passivity of insurance to the incoming climate crisis.

We have found an uneasy alliance in the midway, but a value proposition that seemed a clear cut win and scalable has met the world where it is implemented. That world has teeth.

Solutions that work on paper can often fail in the landscape of the world, or the help that was promised underperforms or is hard to scale.

This is often where the ships crash: most apps that go out to be good and help are often met with the clashing, cold water of our reality. See: National Eating Disorder Association (NEDA)'s chatbot replacing their volunteer staff, Crisis Text Line attempting to diversify revenue by repackaging suicide crisis de-escalation chats as customer service support and getting cancelled, etc.

On top of that, what constitutes acceptable help is constantly changing and shifting - especially given the patchwork funding landscape in the United States. Pushing people to programs like food stamps is a concrete path some years, and others is a brick wall.

This is often sometimes where these companies can fail out. They build a thesis of help, but it's often not-peer reviewed. There is one talented organization called Campbell Collaboration in the non-profit space that goes through and empirically reviews human service programs, and writes thorough and public articles about them.

Other companies have "trials" or research programs that can be available, but are mostly used for marketing programs and internal product development, not development of the greater field or ecosystems around interventions as a whole. These are starts of what the future can be.

This is also why help is just so hard. It is hard to help an individual to the standards that individual needs, but also to the standards of all parties involved as stakeholders in the individual's care. You cannot just build for the user, but the government grants that get them into housing programs, the people in their lives, their caseworkers, their landlords, and everyone else.

Not only do we have to build software that pushes people to a clear definition and point of help, aligned with their incentives, we have to meet it where the world will push it to thrive.

This, in some ways, is where effective altruism genuinely has been innovative in many ways of philanthropy. It is an effective fundraising philosophy strongly matched to the values of the people who can fundraise for it. The "effectiveness" is somewhat dubious depending on who you ask, but it's a philosophy of help that was made within the world, not incubated and then passed onto it.

I would also define harm reduction techniques for drug users as another way where helping people involves meeting them at the level of reality they are at.

Help is also, in many ways, not about being the swooping hero in the night that is able to right all wrongs and reverse all pain. Help is often, simply, being there and measuring the right things, while listening to the right people.

Help can be small, and it can be unmeasurable compared to the traditional way we measure effectiveness of software products. But help that lands in some ways is unmeasurable. It is visible, and knowable to those that have a deep enough knowledge of the community they serve to gain a sense.

Naloxone is an example where the harm reduction community has embraced what is genuinely a miracle: a drug that is able to arrest overdoses and legitimately provide a resurrection. Although not available in all areas and still stigmatized, many organizations I have seen encourage people to carry it and get trained on using it.

CVS Pharmacy by my gym carries it for anyone to buy without a prescription. Naloxone is not reversing the tide of the opioid crisis, but it is a light for each person who knows how to use it and can save a life with it.

A light in the darkness of it all is not the sun, but it never needed to be.

Software needs to help in a deeply nuanced way, and in a way that truly understands what a clear benchmark of help is, all while meeting humans where they are and meeting the software development model as it needs to be in order to be sustainable.

You have to source many different areas to determine what "helped" is, and weigh that with the autonomy of the users and content of the information. You may take scientific research at more of a weight than feedback from users who may struggle with managing emotions or who feel shame.

Truly helping people through software should not just be a result of backwards momentum pushed as the cultural dominance of business values repackaged into philanthropy, but a pragmatic approach that is constantly being tested and debunked.

Help is never this static thing that we should always assume is the end result of our work, it's an end goal that we are always aiming to reach, and that we constantly need to confirm if our work is getting us to that goal.

Despite my pessimism here, there's a lot of companies that I've found to be in the space that give us all north stars. They're scientifically backed, meet people where the problems are, and are making an impact that speaks louder than any daily, or monthly active user count.

Some companies in no particular order that I find impactful: + Sunlight, a communication platform for child welfare cases. Made by a foster parent, and genuinely groundbreaking in potentially reducing the strife and trauma a child and family may go through - https://getsunlight.org/ + Chayn, a platform that has translated gender-based violence tools and trauma informed practices into many different languages for survivors, meeting them where they are in the world. https://www.chayn.co/ + JustFix, a non-profit who builds tools that empower tenants to manage housing rights problems with ease (and who's tools I have used this year when my ceiling collasped). https://www.justfix.org/en/

And many, many more.

I often get so frustrated about the state of "helping" software, mostly because it seems so hard to find pure examples of companies doing their diligence for it instead of applying it as a marketing sheen, but also because it seems even when the work is put in, it is so brittle and fragile that it can crack.

Software that helps people is hard. Perhaps, it is one of the hardest and most gratifying engineering and product development challenges we have to face in our modern world. It is the rarity of achieving that goal that makes it all the more imperative to do so.

Software development is becoming accessible, in good and bad ways, like never before. We have more money being poured into technology and software development, and there is no shortage of AI companies raising millions and setting themselves on fire or burying themselves in fraud every day.

What a failure it would be to have all of these resource and opportunities, and to not seize the moment to give back to our world that gave us these resources.

Like any challenge: this will take time and effort. And if applying software development, a technology that is uniquely high scale and low cost, will not meet the challenges that put us in peril, I don't know what will.

There is, if we can listen and remain within our efforts, an undefinable opportunity to build something inevitable here. I feel it. I've been studying it. I want to build software that genuinely, undeniably helps people for 11 years.

It's all I think about every day, to the point of shaking devotion. It is feasible. It must come. And it needs to, because "software that helps" will forever be a phrase thrown around until we are truly able to build something that stands above everything else in its' commitment and authenticity.