A city spending $690 million a year on homelessness—and still counting
On a single night in January 2024, 8,323 people were counted as homeless in San Francisco. Of those, 4,355 were unsheltered—sleeping in doorways, vehicles, under overpasses, in tents. The city's Department of Homelessness and Supportive Housing runs a $690 million annual budget to address this, a figure 142% higher than it was before the pandemic.
The numbers are moving. Tent encampments hit record lows in late 2025—down 37% from a year prior. More people are getting into shelter. But chronic homelessness ticked up 9% in the last count, and homeless families nearly doubled. The visible is shrinking. The stubborn core remains.
What the numbers don't capture: two-thirds of people experiencing homelessness in San Francisco are people of color. Black San Franciscans make up 5% of the city's population and 25% of its homeless count. This is not random distribution. It is the shape of a century of policy.
Between 2023 and mid-2025, state and city funding connected 7,794 people to services and placed 1,543 into housing. That's real. It's also a fraction of the need. A single unit of subsidized affordable housing in San Francisco costs $730,000 to build. The math is brutal before you even get to the human part.
The people already doing the work, in the rain, tonight
Before any conversation about technology or systems change, there are people already out there meeting the moment. They don't wait for referrals. They go where people are.
Ask anyone working these roles what the minimum viable impact of care looks like. The answer is always the same: dignity. Not a program outcome. Not a statistic. Just—did this person feel seen? Did they get dry? Did someone actually stop?
San Francisco gets outsize attention for its visible homelessness—partly because of geography (density, mild climate, concentrated poverty), partly because of a media ecosystem tuned to it. But the problem isn't local. It's national, and SF is one data point in a much larger picture.
771,000 people on a single night. A count that undercounts.
In January 2024, HUD estimated 771,480 people experienced homelessness across the United States on a single night—the highest recorded number since tracking began. California alone accounts for roughly 187,000 of those, about 28% of the national total despite being 12% of the population.
The reasons people become homeless are rarely singular. In SF's own surveys: job loss (26%), substance use (18%), eviction (13%), asked to leave by family or friends (12%), mental health (8%). These aren't moral failures. They're the predictable results of systems that cut mental health funding by a third between 1978 and 1982 and never rebuilt it, of affordable housing policy that hasn't kept pace with four decades of market pressure.
The United States has a comparatively high rate of street homelessness among wealthy nations—76 per 100,000 living unsheltered, against countries like Finland where the number drops below 20. The US does more in shelter provision than many peers. But the street number reveals something about what the floor looks like here versus elsewhere.
The people who built those ideas weren't short on vision. They were short on infrastructure. In the old world, an idea needed a developer, a server, a budget, a team. Most ideas that emerged from lived experience died quietly because the person who had the idea couldn't get it off the ground. The barrier wasn't imagination. It was everything that came after it.
That wall is gone. Modern AI tools have collapsed the distance between a clear idea and a working prototype to hours. You describe what you want built and it starts taking shape. The question is no longer can this be built? It's who gets to build it now—and for whom?
The global number makes the American number look small—not to diminish it, but to locate it. According to the latest UN-Habitat estimates, 318 million people are homeless worldwide, while 2.8 billion—over a third of the global population—lack access to adequate housing. More than 1.1 billion people live in informal settlements, with 90% concentrated in Africa and Asia.
The comparison between countries is imperfect—definitions differ, data collection differs, what gets counted as "homeless" differs. But the direction is clear. Finland has fewer than 20 homeless people per 100,000. Japan, through coordinated social assistance, has reached record lows. The US, for all its wealth, sits at 76 per 100,000 on the streets alone. These are choices embedded in policy, not laws of nature.
The countries with the worst homelessness figures—Syria, Pakistan, parts of sub-Saharan Africa—are dealing with compound crises: conflict, displacement, climate disaster, collapsed infrastructure. The people of SF are not Syria. The people of the US are not Bangladesh. The gap between what people here face and what people elsewhere face is enormous—and that gap is capacity.
Not a sermon. Just data. We have more room to act than most of the world will ever have.
Homeless per 100,000 population · OECD comparison
SF figure reflects its density and geography as an outlier within the US. OECD figures via Our World in Data / OECD Affordable Housing Database.
Built under deadline, with strangers, for people who needed it
These projects exist because a group of people got in a room with a real problem, a deadline, and enough care to ship something. None of them are finished. All of them are proof that the distance between an idea and a working thing has genuinely collapsed—and that the most important applications haven't been built yet.
PathfinderXR
An XR navigation tool built for people navigating reentry—helping them find services, build spatial confidence, and reclaim a sense of place after incarceration. The city as something navigable, not hostile.
→ devpost.com/software/pathfinderxrChorus
A care coordination platform that reframes the elder from passive recipient to active mentor—matching caregivers, seekers, and guides through narrative compatibility rather than demographics. Built at AgeTech SF with a team of strangers in 48 hours.
Pilot data: 43% reduction in elder loneliness. 38% reduction in caregiver stress.
🏆 Prize Winner · AgeTech SF HackathonROOM
A multi-ontology coordination layer for human-AI collaboration. Built on the premise that knowledge lives in context, not containers—and that the most important coordination happens between people and agents with genuinely different ways of knowing.
🏆 1st Place · PLACES · RP1 Open Metaverse HackathonCare Is Currency — The Tokenomics of Care
The HOT worker who shows up at 2am. The ambassador who de-escalates with nothing but presence. Every one of those acts generates value that disappears from every ledger we have because we have no way to record it.
The tokenomics of care isn't about monetizing human connection. It's about making the invisible exchange legible. The elder who shares lived wisdom is giving something real. The returning citizen who mentors someone just coming out is generating social capital the system never credits back. What gets measured gets valued. What gets valued gets resourced.
Care isn't a deficit to be managed. It's an economy to be designed.
Who's already doing the work—and where the gap shows up
These organizations are active, serving real people now. The question isn't whether they need better tools. It's whether tech is going to show up as another extractive layer—or as something that listens to what their communities already know and amplifies what's working.
Workforce readiness and upward mobility for people locked out of economic participation. The gap: people learn tools but rarely get to build something real. The opportunity: close the loop between learning and making so participants leave with something they built, not just a credential they earned.
Street-level transformation through dignity and presence. Their ambassadors carry extraordinary lived knowledge that no database captures. The gap: no tools to translate that knowledge into leverage—story, systems, income. The opportunity: help people document, narrate, and build real pathways from what they already know.
Real coding education in SF's most complex neighborhood. The gap: the curriculum hasn't caught up to the moment. Agentic tooling and AI-native workflows are a faster, more immediate on-ramp—and more accessible for people who never thought of themselves as technical.
Food, shelter, recovery. Meeting people at their most urgent point of need. The gap: survival is not a destination. The opportunity: introduce pathways toward creative and economic agency—not as an add-on, but woven into how recovery is designed from the start.
Two asks. Both honest. Neither vague.
This dispatch has two audiences. Program managers, counselors, advocates, and ambassadors working inside these ecosystems—people who have the trust of a community and are looking for practical tools to bring in. And builders: people who see the same gap and want to do something about it, but don't know where to plug in.
A 90-minute session where participants with zero technical background build something real—a tool, an artifact, something they can use, show, or evolve. That's how you find out what's actually useful in a given context. Not by theorizing. By doing it once.
If you're a builder—someone who codes, facilitates, designs, teaches, or just shows up consistently—the goal isn't one person running workshops forever. It's a network: people who each plug into the communities they're already close to, with tools that actually work.
The bigger ask: let's figure out what this looks like as a model—something that scales because it's true, not because it's funded.
I run programs. Let's test this.
You have the community. The session design is ready. 90 minutes, one outcome—real people leave with something they made.
Let's run a pilot →I'm a builder. I want in.
You code, design, facilitate, or teach. You see the same gap. Let's find where skills and communities overlap.
Let's connect →