> ## Content Index
> Fetch the complete content index at: https://www.vitalcitynyc.org/llms.txt
> Use this file to discover other available public pages before exploring further.

# AI to Help Lift People Up
- URL: https://www.vitalcitynyc.org/ai-reentry-job-coaching-criminal-justice/
- Published: 2026-11-18T16:45:00.000Z
- Updated: 2026-11-18T17:10:44.000Z
- Description: How technology helps with reentry — and can assist with other complicated social services work
- Author: Sam Schaeffer
- Tags: #ai-and-the-city, Artifical Intelligence, Incarceration, Technology, Better Prompts, Better Services

Travis came to the Center for Employment Opportunities (CEO), the nonprofit I lead, just weeks after being released from prison. For months, he worked hard to get ready for his new chance. He worked on a transitional job site, consistently showed up for job coaching sessions and practiced diligently in mock interviews. Job interviews can be high-stakes for any of us, but for someone with a criminal conviction — who’d last had a job six years ago — Travis had a pit in his stomach. 

Travis is just one of the 600,000 people returning from prison nationwide every year with few, if any, reentry resources. The cycle of poverty and recidivism — costing billions — is nota product of individual moral failings. Recidivism is complex and is often the result of generations of underinvestment by systems that were built on bias. For someone like Travis, who did everything right to prepare himself for a job, a little extra help on the nights and weekends felt necessary to nail a job interview and calm his nerves. But what traditional service provider offers that kind of around-the-clock outreach? 

It turned out, AI — often pilloried as a force that’s threatening people’s livelihoods — could help. The night before Travis’s first interview, he dialed a number from the comfort of his own home. On the other end was CoachLink, an AI voice coach ready to run through as many mock interviews as Travis needed to feel confident and capable of providing immediate feedback.

Travis went to the interview. He got a job with an HVAC company. He did really well for nine months. But like so many of our participants, the story doesn’t end there. His hours were cut back, and he was eventually laid off. At that point, AI couldn’t help. Our staff brought Travis back into the organization’s care, helped him meet basic needs, and lined up new interviews for him. 

Travis’s story clearly illustrates the need for dual and symbiotic investments in people and supplemental AI tools that target core capacity problems faced by mission-driven service organizations. 

To that end, we’ve drawn up a framework for direct service providers like us to play a key role in developing effective and responsible AI. We think organizations should incorporate four essential lessons into their AI strategy.

First, we must meet people where they are by making tools accessible through familiar mechanisms — not expect them to own a laptop or download an app. Second, AI should never seek to replace human support, which is needed to help people grappling with complex life challenges, but instead supplement it. AI’s ability to quickly expand reach is not an excuse to neglect investment in people. Third, practitioners must look everywhere for unintended consequences. Service providers must assess any potential harm and guarantee efficacy. Finally, service providers, organizations with deep expertise, and the people being served must come together to lead AI tool development in the social sector. 

If these guidelines are followed, we can build truly useful technological tools that help lift people up. I’ve seen it firsthand.

### Meet people where they are

At the beginning of CEO’s AI journey, we considered building an app. Based on feedback, we abandoned the idea. In our experience working side by side with people leaving incarceration, asking someone to download an app, create a profile and manage a password results in a 75% drop in engagement. Furthermore, many people returning from prison have smartphones with limited capabilities, and they experience inconsistent Wi-Fi access; the same can be said of many people struggling to improve their lives.

So instead of an app, we applied AI to a traditional and familiar communication apparatus: the telephone. Phones meet people where they are. Indeed, in this case phones let us start using tools even before people were released. Prisons and jails don’t allow internet access, but they do have phones, and tablets are becoming pervasive within prison walls.

[CoachLink](https://www.ceoworks.org/ai-for-opportunity?ref=vitalcitynyc.org), CEO’s voice coach is already showing a strong impact. An internal pilot involving 500 CEO participants showed a 15% increase in job placements. Just think of how telephone coaching could relieve the workload of dedicated but overburdened coaches while also helping them develop hard and soft skills in their own careers.

> AI’s capacity cannot become an excuse to underinvest in the human relationships that drive lasting change.

### Technology’s potential isn’t a cheat code to scale impact. AI should supplement, not replace

I’ve been struck by the scale technology can achieve with relatively little friction. One of our prison-based technology initiatives — a job-training course available through tablets has reached over 400,000 incarcerated learners across the U.S. This kind of success entices both the public and private sectors to use AI as a shortcut for reaching tens of thousands of people at low cost. 

Though tempting, AI’s capacity cannot become an excuse to underinvest in the human relationships that drive lasting change for people facing complicated challenges. The reentry field must continue to push for sustained, predictable funding that enables organizations to hire, train and retain the people who make successful reentry possible.

This is true for all sorts of human-service providers. For example, career navigation apps that help people identify the availability of training programs can be helpful, but the app only takes someone so far. To develop the hard or soft skills necessary to succeed in a certain profession requires human instruction. And that human instruction requires investment and protection.

Widening inequality coupled with a rupture in the social safety net at the federal level has exponentially amplified an age-old struggle for service organizations: People need more support than we can offer with our limited budgets. AI won’t solve a capacity crisis. It can, however, address voids in services and fundamentally expand human relationships *if* the social sector leads its development.

The work of our job coaches is essential. They represent the compassionate, patient and crucial human touchpoint for people returning from incarceration. Sometimes, they are the only form of support in a returning individual’s life. That is a heavy responsibility. Their day-to-day is stressful, causing burnout and disengagement. 

We created an AI voice tool not to replace our job coaches, but to support staff in some of the most difficult parts of their jobs. With this tool, our staff can rehearse de-escalation or workplace conflict scenarios. Our motivational interviewing training — a key practice in developing behavior change — produced a 10% improvement in competency scoring in as little as an hour with constructive feedback from the AI tool. 

> Could a tool designed to improve preparation inadvertently make some decision-makers more skeptical, leading to worse outcomes for the very people it is intended to help?

### Look for unintended consequences everywhere

Tech companies don’t mind failing as long as they fail fast — and try again. Working in the justice system, we don’t have that luxury, because the stakes are someone’s freedom, livelihood or safety. Even interventions that seem obviously beneficial can produce unintended consequences once they interact with complex systems.

Here’s an example. Recently, we partnered with Project Commutation, an Oklahoma nonprofit, to build a voice coach that prepares people for parole hearings. 

Every year, thousands of people appear before parole boards without the benefit of legal counsel. Imagine sitting in front of an intimidating panel with your liberty on the line. You may feel anxious and struggle to articulate a vision for a positive future. A parole preparation voice coach, we thought, could offer someone the opportunity to practice telling their story and answering common questions.

At first glance, the idea seemed unquestionably positive. I’ve heard directly from parole board members that they want to see someone put their best foot forward. But what if board members began to notice similar phrasing among applicants and assumed they used AI? That’s already happening with AI usage in writing. If you didn’t get the preparation right, could you emphasize the wrong issues in your hearing? Could a tool designed to improve preparation inadvertently make some decision-makers more skeptical, leading to worse outcomes for the very people it is intended to help?

Before deploying the tool, we are engaging attorneys, formerly and currently serving parole officials, and people preparing for hearings to fully understand how it could influence the entire system — not just the individual using it. 

Human services organizations should bring the same mindset to every single AI application. The question is not simply whether a tool works for its user. It is whether it changes incentives, behaviors or decisions elsewhere in the system in ways that could unintentionally cause harm. 

### Build together

Last April, a group of local criminal justice reform nonprofit leaders — including those who run programs for women leaving incarceration, jail diversion initiatives, advanced training for returning individuals and parole preparation — convened to discuss AI. At the start of the day, leaders expressed excitement about the prospect of integrating AI into their initiatives. As morning stretched into afternoon, the largest obstacle became clear: on their own, these organizations lacked the technological capacity and overall bandwidth to explore AI usage.

There were no native technologists in the room. The ideas were all generated by people with close proximity to the justice system or experience within the justice system. 

Rather than build in a vacuum within one organization, the future of AI in human services will benefit from cooperation, both within organizations and between them. The front-desk idea — proposed by overwhelmed people who do a tough job every day — was immediately embraced by several organizations that were facing the same challenge. Instead of having each organization develop something independently, working together allows us to pool technical capacity while drawing on a broader range of practitioner and lived experience. Building AI tools collectively allows our sector to build, test and continually improve tools while identifying risks. 

Put differently, nonprofit practitioners, more than technologists, should be at the center of AI tool development. They and the people they serve are best positioned to assess whether an AI can do more good than harm.

Not enough direct service providers are building the tools we need to help the people we serve. We can’t wait for others to create technology that may harm individuals coming through our front doors. I believe we are in a unique position to deploy these tools infused by our human-centered work, so that people like Travis can truly benefit from what AI has to offer.