Street Smart: How Street Medicine Is Proving AI Works for Healthcare's Hardest Cases
- Mar 17
- 13 min read
Updated: 7 days ago
May 2026 Authors: Karthik Murali, PhD, Akido Labs Dr. Rishi Patel, MD, Akido Labs Jared Goodner, Akido Labs Prashant Samant, Akido Labs
Executive Summary
American healthcare faces a moment of unprecedented convergence. Medicaid cuts threaten to unravel safety-net care across the country. A physician shortage crisis deepens, with projections showing a deficit of up to 86,000 doctors by 2036.(1) And artificial intelligence has matured from experimental curiosity to clinical reality—with 66% of U.S. physicians now using AI tools, up from 38% just two years ago.(2)
The question is no longer whether AI will transform healthcare delivery, but how it will do so in the real world. Will it simply automate documentation—or will it force-multiply clinical teams and expand true care capacity, enabling providers to reach more patients without sacrificing quality?
Key Findings
53% of patients see a medical provider on their very first day of engagement
82% / 63% retention at three and six months, respectively—in a population where dropout is typically the norm
2.7 visits/month average across the enrolled population; 4.5 visits/month for high-acuity patients
~2,400 patients treated for substance use disorder, with 83% initiated on Medication-Assisted Treatment (MAT)—the gold standard for addiction care
Akido patients, on average, use the emergency department (ED) 1.4 times a year, a 55% reduction compared to the national rate of 3.1 ED visits a year.
100% of program costs covered by Medi-Cal reimbursements
150 (avg street med panel size) → 500 provider panel size progression; target of 1,000 as AI integration deepens
These outcomes were not achieved by pushing teams to burnout or unsustainable resource expenditure. They were achieved by amplifying the work of individuals through systemic integration of medical intelligence and automation technology through the care delivery process—a program designed, from the ground up, to intertwine technology and clinical care.
The implications extend well beyond street medicine. For health systems navigating Medicaid cuts, physician shortages, and value-based care mandates, the Los Angeles program offers a proof of principle: that AI-powered care can simultaneously improve access, engagement, and clinical outcomes—and do so sustainably.
I. The Problem: Safety-Net Healthcare Is Under Pressure
American safety-net healthcare is navigating a convergence of structural forces that, taken together, constitute an inflection point. Medicaid funding faces proposed cuts of historic scale. A primary care workforce already stretched thin is projected to fall 10,000 providers short in California alone by 2030. And the populations most dependent on both—those with the fewest resources, the most complex medical needs, and the least access to alternatives—have nowhere else to turn.
Nowhere is this pressure more acute than among people experiencing homelessness. In California, more than 200,000 people are living unsheltered—a number that continues to rise. And the clinical picture of that population tells its own story.
People experiencing homelessness (PEH) die, on average, approximately 30 years earlier than their housed peers.(3) Forty-four percent suffer from substance use disorders. Sixty-seven percent live with mental health disorders.4 They carry burdens of untreated chronic disease, infectious disease, trauma, and malnutrition—conditions that are, in most cases, entirely treatable. The existence of treatment is there. What they lack is a healthcare system capable of reaching them.
These patients are not falling through cracks in the system. They were never inside it.
As Medicaid faces deep structural cuts, even the thin safety net that currently exists for this population is at risk. There is no doubt that the problem is real. The question is whether there is a care delivery model capable of meeting it.
II. The Known Solution: Street Medicine Works
Against this backdrop, one approach has consistently demonstrated it can reach patients that the traditional system cannot—and change their outcomes.
Street medicine—delivering care to patients wherever they are, on their terms, without the bureaucratic preconditions of the traditional clinic—removes the barriers that prevent the most vulnerable patients from ever receiving care. The most effective programs don’t just meet patients physically where they are. They bring the full range of services those patients need: primary care, addiction medicine, psychiatric services, and intensive case management, integrated into a single encounter. A 2025 study found that 93% of unsheltered individuals in Los Angeles County had no access to primary care—the population street medicine exists to reach.(6)
Programs like USC Street Medicine and Healthcare in Action (HIA) have demonstrated that the model produces real, measurable results. USC Street Medicine reports a 75% reduction in hospital admissions among the individuals it has served, a decrease in average length of hospital stay from 12 to 7.9 days, and a 42% housing placement rate within one year of first engagement.(7) HIA, which delivers integrated medical, behavioral health, and housing navigation services across six California counties, has similarly demonstrated the model’s capacity to reach patients living entirely outside the traditional care system.(8) A peer-reviewed evaluation of a street medicine program serving unsheltered individuals confirmed these patterns, documenting meaningful improvements in healthcare outcomes through a street-based model of care.(9)
Meeting patients where they are, with integrated services, produces better outcomes than waiting for them to appear at a clinic they will never walk into. The evidence base is no longer in dispute. What is in dispute is whether it can move beyond its current scale.
III. The Blocker: Street Medicine Can't Scale
Despite its demonstrated effectiveness, street medicine has remained small, fragmented, and grant-dependent. Three structural constraints have kept it that way.
The Charity Trap
Most street medicine programs run on philanthropic funding and volunteer labor. This patchwork model lacks the financial consistency and predictability required to treat street medicine as a true medical specialty. Programs can’t hire, can’t retain, and can’t plan. When grants run out, programs shrink or close—regardless of outcomes.
The Provider Shortage
Street medicine, in its traditional form, is a provider-heavy model. Panel sizes of 100 to 200 patients per physician are common—a fraction of what a primary care provider manages in a standard clinic setting. Against a need population of 200,000 unsheltered individuals in California alone, the arithmetic is unworkable. Serving them at traditional panel sizes would require 1,000 to 2,000 street medicine providers that, in a state already facing a projected shortfall of 10,500 primary care providers by 2030,5 simply do not exist.
The Math Problem
The economics of street medicine at current scale are inherently unsustainable. Low panel sizes mean high per-patient costs. Every encounter requires significant clinician time for intake, documentation, care coordination, and follow-up. Without a structural mechanism to multiply provider capacity, the cost per patient can’t come down—and the program can’t survive without ongoing subsidy.
The result is a healthcare model that has proven it works—but can only serve hundreds of patients when the need is in the hundreds of thousands. Programs remain heroic, small-scale efforts. Excellent. Undeniable. And structurally incapable of reaching the problem they exist to solve.
IV. The Moment: AI Has Matured Enough to Change the Equation
AI in healthcare has crossed from experimental to operational. Sixty-six percent of U.S. physicians now report using AI tools in their practice—up from 38% just two years ago.(2) That is a structural change in how medicine is being practiced.
But while the rise in adoption rates is significant, the more important factor is the AI tools’ capabilities. The current generation of clinical AI is no longer limited to automating documentation or flagging billing codes. It is capable of guiding comprehensive patient intakes, surfacing relevant clinical information in real time, supporting diagnostic reasoning across specialties, and generating structured assessments for provider review—all in field conditions, on a tablet, in the hands of a non-physician care team member.
This matters because it changes the fundamental constraint of street medicine. The bottleneck isn’t the care itself. It is the ratio of licensed providers to the patients who need them. If AI can extend what a single provider can supervise—by enabling community health workers and medical assistants to conduct comprehensive, clinically-guided encounters—then the panel sizes that have made street medicine unscalable become, for the first time, a solvable problem.
The convergence is now: the problem is worsening, the solution is proven but stuck, and the enabling technology has arrived.
V. Akido's Answer: Street Medicine Intertwined with AI
When Akido set out to build a street medicine program in Los Angeles, the goal was to build something that could reach the 200,000 people experiencing homelessness who need medical care across California—and to prove that AI-enabled care is more than a supplement to the traditional model–that it is, in fact, a structural replacement for the traditional model’s most limiting assumptions.
Akido picked the hardest possible place deliberately. This is where the traditional system has no sustainable answer, and where AI-enabled care has to show real impact if it intends to credibly address the broader crisis facing American safety-net healthcare.
The design premise was simple but consequential: a program where technology and clinical care are intertwined at every level, from the first patient encounter to provider oversight to operational scaling.
Three things make this possible.
An Irreplaceable Human Core
The program runs on a team of community health workers, field medical assistants, substance abuse counselors, nurses, and street medicine providers who are, by any measure, exceptional. They were not recruited as representatives of a technology demonstration—but because care delivered in the field, to patients who have every reason to be wary, depends entirely on the quality of the human relationship. Without that foundation, no technology matters. It is the precondition.
Powerful, Purpose-Built Technology
The second component is a combination of medical intelligence and automation technology—built for field conditions. Akido’s medical intelligence, ScopeAI, powers a tech stack designed to support the full arc of a patient visit: guiding intake and investigation, surfacing relevant clinical information in real time, and generating structured assessments for rapid licensed provider review. It gives community health workers the clinical scaffolding to conduct encounters across primary care, addiction medicine, and behavioral health—without requiring a provider to be physically present for every interaction. All clinical assessments and treatment decisions remain with licensed providers, who review before any diagnosis or treatment is initiated. Providers are core to the success of the program, and the medical intelligence in ScopeAI enables them to operate beyond the limits of what the traditional model permits.
A Design Process That Intertwines Them
The third element is the least visible and perhaps the most important. Technology integration in healthcare fails predictably when it is deployed all at once—when a powerful tool is introduced into a clinical environment without the iterative, workflow-level process required for clinicians to trust it and use it well.
Akido’s approach was different. Start with a working clinical program. Start with capable technology. Then systematically integrate them, one workflow at a time: test, gather clinician feedback, adjust, move to the next. Each cycle expanded the reach and capacity of the same care team—first doubling it, then tripling it, then quadrupling it—while the number of meaningful clinical encounters per patient actually went up. Instead of diluting the care, the technology concentrated it.
Outcomes
Since launching in March 2023, Akido’s LA street medicine program has served approximately 7,000 patients and enrolled 1,800 into ongoing comprehensive care.
Scale & Access
53% of patients see a medical provider on their very first day of engagement
7,000+ patients served since launch; 1,800 enrolled in ongoing comprehensive care
Engagement & Retention
82% retention at three months
63% retention at six months
2.7 visits/month average across the enrolled population; 4.5 visits/month for high-acuity patients
Clinical Outcomes
~2,400 patients treated for substance use disorder, with 83% initiated on Medication-Assisted Treatment (MAT)—the gold standard for addiction care1
363 patients treated for psychiatric disorders
Akido patients, on average, use the ED 1.4 times a year, a 55% reduction compared to the national rate of 3.1 ED visits a year.
Financial Stability
100% of program costs covered by Medi-Cal reimbursements, with no philanthropic grants required
150 → 350 → 500 provider panel size progression coupled with 80% positive patient response; target of 1,000 as AI integration deepens.
30% increase in case manager panel sizes
VI. The Design Process: How You Intertwine AI and Clinical Care Without Breaking Things
The iterative integration process is the structural reason programs like this succeed—and the reason others fail.
The Cautionary Tale
IBM Watson Health spent billions and promised to revolutionize oncology with AI. It failed because while the technology was powerful, it was also deployed wholesale into clinical environments where workflows hadn’t been redesigned around it. Clinicians didn’t trust it. The AI wasn’t grounded in real field conditions. The gap between what the technology could theoretically do and what it could actually do in practice—on a given day, with a given patient, by a given care team member—was never bridged.
Technology that is not integrated into real clinical workflows through a deliberate process gets rejected, no matter how powerful it is. The technology wasn’t the problem, the implementation was.
How Akido’s Process Works
The method is straightforward, but the discipline required to execute it is nuanced. Start with a great clinical program—one that already produces outcomes, already has a care team that trusts each other, already has workflows that function. Then you earn the right to integrate technology by having a program worth integrating it into.
Start with capable technology built for field conditions. Third-party solutions typically cannot be deployed in a street medicine context. The infrastructure has to be your own, or you cannot iterate on it quickly enough.
Then integrate one workflow at a time. Test. Get clinician feedback. Adjust. Move to the next. The process is continuous, not episodic. With each cycle the technology improves and so does the capacity and reach of the care team as a whole.
What This Looks Like in Practice
Two examples from Akido’s current program illustrate the difference the design process makes.
Example 1: AI-Guided SUD Intake and Investigation
Before: A community health worker in the field is talking with someone struggling with fentanyl—someone who, in this particular moment, is open to getting help. Connecting them to treatment meant connecting them to a street medicine provider who could prescribe Suboxone and start medical-assisted treatment (MAT). Best case: the provider was available that day. More often: scheduling constraints pushed the appointment to the next day, or the day after.
That gap is where patients are lost. For someone in active addiction, living unsheltered, the window of willingness is fragile. They go back to using. The moment passes..
After: The patient calls Akido’s MAT access line directly. A substance abuse counselor—a non-prescriber—conducts a full MAT intake right there, in the field, in the moment the patient is ready. AI-guided tools walk the counselor through the encounter: investigating and prompting the right questions, capturing the right information, generating a structured summary for review. A licensed provider reviews the completed intake with the patient in five to ten minutes, instead of conducting a full thirty-minute intake themselves and then completes their clinical assessment and makes the prescribing decision themselves. The patient receives a Suboxone prescription within a median of 18 hours—compared to days under the previous model.
The counselor could not do this without the AI. The AI could not do this without the counselor. Together, they capture the moment when the patient is ready—and that moment is everything..
Example 2: Field Medical Assistants Reaching Remote Populations
Before: People in remote parts of the county had near-zero access to timely, high-quality care. Under the traditional model, a provider and medical assistant might reach a remote area once a month—or once every six months. Patients who weren’t there that day, or weren’t ready that day, didn’t get seen. By the time the team returned, the window had closed.
After: Field medical assistants are deployed in remote areas with tablets running AI-guided intake and investigation tools. They conduct comprehensive medical intakes—history, vitals, screenings, medication reconciliation—without requiring a provider to be physically present. The provider reviews intake information remotely and addresses a significant percentage of visits immediately. Real medical needs are solved in the field, in real time—not deferred to a follow-up that may never happen.
Three things are achieved simultaneously: capturing the patient in the short window when they are ready and willing, delivering a real clinical touchpoint that builds trust, and extending provider reach to areas that would otherwise go months without coverage. Only approximately two in five cases require deep, synchronous provider involvement—the same provider team covers dramatically more ground.
VII. Conclusion: The Model Works. Now Scale It.
We needed an exceptional care team to create the outcomes documented in this paper. But exceptional teams alone cannot reach the 200,000 unhoused individuals across California, or the millions more living on the edges of the safety net. There is provider scarcity. There is never enough funding. The traditional model cannot scale to meet a problem of this magnitude.
AI is the force multiplier that changes the equation—the mechanism capable of taking what works for 1,800 enrolled patients and making it work for the next 5,000, and the 5,000 after that, and for the patients in the furthest reaches of the county where no provider has the capacity to go today.
The path forward will still require heroic effort and a systematic replication of a model that has been proven to work—built on a design process that integrates technology into care deliberately, iteratively, and without breaking the human relationships that make it possible in the first place.
What works for America’s most vulnerable patients—people experiencing homelessness, active addiction, serious mental illness, and complete disconnection from traditional healthcare—can work in multiple environments. The streets of Los Angeles are far from the exception. They are the proof of concept.
The same medical intelligence and automation technology that enables a community health worker to conduct a MAT intake in a parking lot can enable a primary care provider in a rural clinic to practice at specialist-level depth. It can help a safety-net health system absorb the impact of Medicaid cuts without cutting access. It can give a Medicaid managed care plan a scalable infrastructure for the high-complexity, high-cost populations that traditional care management models have never been able to reach. The care delivery challenges facing street medicine—too few providers, too many patients, too little time, too much administrative burden—are the same challenges facing the broader healthcare system. What Akido has built in Los Angeles is more than a street medicine program that happens to use AI. It is a template for what AI-enabled care looks like at scale, starting from the hardest possible place.
Health systems, managed care plans, policymakers, and provider organizations looking for a model that works— in sustained practice with the most complex patients—will find one here.
References
1. Association of American Medical Colleges. "The Complexities of Physician Supply and Demand: Projections From 2021 to 2036." AAMC, 2023.
2. OpenAI. "AI as a Healthcare Ally: How Americans Are Navigating the System with ChatGPT." January 2026. [Physician AI adoption figures sourced from AMA survey data cited therein.]
3. United States Interagency Council on Homelessness. "Homelessness Data & Trends." USICH, 2024. https://usich.gov/guidance-reports-data/data-trends
4. Wiens K, et al. "Prevalence of Mental Health Disorders Among Individuals Experiencing Homelessness: A Systematic Review and Meta-Analysis." JAMA Network Open. 2024. https://pubmed.ncbi.nlm.nih.gov/38630486/
5. California Health Care Foundation. "Five Ways to Cure California's Doctor Shortage." CHCF. https://www.chcf.org/resource/cure-californias-doctor-shortage/ [Citing UCSF Healthforce Center estimate of 10,500 additional primary care providers needed by 2030.]
6. Coulourides Kogan A, Feldman B, Feldman CT, et al. "Access to Basic Needs and Healthcare by People Experiencing Unsheltered Homelessness." Journal of Primary Care and Community Health. 2025. https://journals.sagepub.com/doi/10.1177/21501319251356768
7. Keck School of Medicine of USC. Street Medicine Program Outcomes. USC Street Medicine, 2024. https://keck.usc.edu/street-medicine/about/ [Note: figures reflect USC Street Medicine's internal program reporting, not a peer-reviewed publication.]
8. Healthcare in Action. "What We Learned After Four Years of Offering Street Medicine." STAT News, January 24, 2025. https://www.statnews.com/2025/01/24/homelessness-street-medicine-medical-van-scan-group-healthcare-in-action/
9. Feldman BJ, Kim JS, Mosqueda L, et al. "From the Hospital to the Streets: Bringing Care to the Unsheltered Homeless in Los Angeles." Healthcare (Amsterdam). 2021;9(3):100557. https://pmc.ncbi.nlm.nih.gov/articles/PMC7800740/
10. Sordo L, et al. "Mortality risk during and after opioid substitution treatment: systematic review and meta-analysis of cohort studies." BMJ. 2017;357:j1550.
11. Centers for Disease Control and Prevention. "Emergency Department Visits Among Persons Experiencing Homelessness." MMWR Morb Mortal Wkly Rep. 2023;72(42). https://www.cdc.gov/mmwr/volumes/72/wr/mm7242a6.htm
For more information about Akido’s programs and technology platform, contact Akido Labs.