Thursday, August 6
Business · Technology · Leadership

The Neurodiversity Decisions a Founder Makes Before HR Exists

Key Takeaways

  • Most workplace neurodiversity failures happen in hiring design, not in hiring intent. The interview format screens for interview performance rather than job performance.
  • Job descriptions written in traits rather than tasks act as filters against candidates who could do the work.
  • Accommodations that must be requested carry a disclosure cost, which is why default accommodations reach more people than available ones.
  • Neurodivergent employees frequently need more structure rather than more flexibility, which is the opposite of how most inclusion policies are written.
  • In a company small enough that the founder still designs the work, these are founder decisions rather than HR decisions, and they are cheapest to make early.
  • Dr. Deena Moustafa runs a multi-state clinical business and a direct-to-consumer apparel brand off one team, which has made the operational version of these questions unavoidable rather than theoretical.

Dr. Deena Moustafa is a Board Certified Behavior Analyst, an associate professor, and the founder of two companies that have almost nothing in common on paper.

The first, Go Behavioral, delivers behavioural health services for children and young people with autism and ADHD across California and Florida, in homes, schools, clinics, community settings, and virtually. The second, Cozyino, is a direct-to-consumer apparel brand making sensory-friendly and adaptive clothing, sold online at consumer price points.

One is a clinical services organisation operating under payer contracts and licensure requirements. The other is a retail business competing on product, margin, and fulfilment. They share a founder, largely one team, and a multilingual staff working across two continents.

Running both has forced a set of operational questions that most founders encounter years later, if at all, and that Moustafa argues are considerably cheaper to answer at the beginning.

The Interview Is The First Design Decision

Her starting position is that the standard hiring process does not measure what employers think it measures.

A conventional interview is an unstructured verbal exchange under time pressure with an unfamiliar person, assessed partly on eye contact, conversational rhythm and the ability to improvise a narrative. Almost none of those are requirements of most jobs. They are requirements of interviews.

The practical consequence is that a process designed to identify capability instead identifies fluency in a particular social format, and screens out candidates whose capability is real but whose fluency in that format is not. The result reads to the employer as a shortage of qualified applicants.

Speaking to Insights Care about entry into her own field, she has framed the first requirement as establishing a strong educational foundation in behaviour analysis. The observation cuts both ways. A sector that asks candidates to build that foundation and then screens them on conversational fluency is discarding the thing it asked for.

Her fixes are unglamorous and cost nothing. Send the questions in advance, which improves the quality of every answer from every candidate and removes an arbitrary memory task. Replace the hypothetical question with a work sample. State the format and duration of the interview beforehand, including who will be in the room. Allow written responses where the job itself is written.

None of this is an accommodation in the legal sense. It is a better assessment, and it happens to remove a barrier.

Traits Are Filters, Tasks Are Requirements

The second decision sits upstream, in how roles are written.

Job descriptions across most sectors are assembled from trait language. Strong communicator. Thrives in a fast-paced environment. Comfortable with ambiguity. Culture fit. Each of these reads as a requirement and functions as a filter, and in most cases none of them describes anything the job actually involves.

Moustafa’s argument is that a description written as tasks rather than dispositions is both more useful to the hiring manager and dramatically more accessible. A role that requires someone to write clear documentation should say that. A role that requires someone to handle unpredictable interruptions should say that too, because a candidate for whom that is genuinely difficult deserves to know before applying, and a candidate for whom it is fine will not be filtered out by a proxy.

The trait-language habit is also where a small company inherits assumptions from larger ones without examining them, usually by copying a template.

Default Beats Available

The most consequential distinction she draws is between accommodations that exist and accommodations that must be asked for.

An employee who has to request a change has to first disclose something about themselves to a manager, early in a working relationship, with no guarantee about how the information travels. That disclosure has a cost, and the cost falls hardest on the people least confident of their standing, which is usually the newest and most junior employees.

Anything that can be made a default therefore reaches more people than the same thing offered on request. Agendas circulated before meetings. Written summaries after them. Decisions recorded rather than communicated verbally in passing. Predictable start times. A quiet space that is bookable by anyone for any reason and requires no explanation.

Every item on that list is also, straightforwardly, better management. That is the argument, and it is a stronger one than the moral case because it does not depend on anyone’s goodwill surviving a difficult quarter.

Structure Is The Accommodation, Not Flexibility

The point she makes that most often surprises other founders is that inclusive workplace design is frequently written backwards.

The default assumption is that neurodivergent employees need flexibility. Often what they need is the opposite: clearer expectations, explicit success criteria, defined scope, and communication that does not rely on inference. Ambiguity is not freedom when the unwritten rules are the hard part.

A workplace that says “we’re flexible, just figure out what works” has not removed a barrier. It has moved the work of figuring out the rules onto the employee and made it invisible, and it will then read the resulting struggle as a performance problem.

Specificity costs a manager very little, and it is the single most portable practice in this list.

Why This Lands With Founders Specifically

The reason Moustafa frames these as founder decisions rather than HR decisions is structural. In a company small enough that the person who designs the work is also the person doing the hiring, these defaults are still choices. Once an organisation is large enough to have inherited processes, changing them becomes a policy project with a budget and a timeline.

The window in which a founder can set them by simply deciding is short, and it closes quietly.

There is also a reason this argument tends to arrive through women founders more often than the reverse. Moustafa’s own clinical interest has long included women on the autism spectrum, and the pattern she describes there is late identification, sustained masking, and difficulties that get read as personality rather than as unmet need. That pattern does not stop at the office door. A leader who recognises it is more likely to look at a workplace and see design decisions where others see individual shortcomings.

The autism prevalence figures underline the scale. The Centers for Disease Control and Prevention’s most recent estimate, published in April 2025 and covering surveillance year 2022, put prevalence among 8-year-olds at 1 in 31. Those children are entering the workforce over the next decade and a half, and they will be hired by companies that are being designed now.

The founders making those design decisions are, in most cases, not thinking about them at all.

Frequently Asked Questions

What Is The Common Mistake Companies Make On Neurodiversity Hiring?

Treating it as an intent problem rather than a design problem. The interview format itself screens for social fluency rather than job capability, and no amount of goodwill in the room corrects for that.

How Much Do These Changes Cost?

Most cost nothing. Sending interview questions in advance, writing job descriptions as tasks, circulating agendas, and recording decisions are process changes rather than budget items.

Why Default Accommodations Rather Than Accommodations On Request?

Because requesting one requires disclosing something personal to a manager early in a working relationship. That cost falls hardest on the newest and most junior employees, so an available accommodation reaches fewer people than a default one.

Do Neurodivergent Employees Need More Flexibility?

Frequently, the opposite. Clear expectations, explicit success criteria, and defined scope tend to matter more than loose arrangements, because ambiguity shifts the work of decoding unwritten rules onto the employee.

Who Is Dr. Deena Moustafa?

A Board Certified Behavior Analyst and associate professor, and the founder of Go Behavioral, a behavioural health provider operating in California and Florida, and Cozyino, a sensory-friendly and adaptive apparel brand. She is also the author of four books published between 2009 and 2011, including You Are Not Alone: A Message to Parents of Children With Autism.

Disclaimer: This article is provided for general informational purposes only and does not constitute legal, medical, employment, or human resources advice. Workplace requirements and accommodation obligations vary by jurisdiction and individual circumstances. Employers and employees should consult qualified professionals regarding their specific needs and responsibilities.

Resect AI Launches With $25M to Tackle AI Hallucinations

Resect AI emerged from stealth on September 3 with $25 million in funding to develop technology aimed at detecting and modifying problematic AI model behavior in real time. The Washougal, Washington startup is targeting enterprise concerns around hallucinations, governance, and oversight as businesses expand their use of generative AI.

Key Takeaways

  • Resect AI launched from stealth on September 3, 2026, with $25 million from private-equity investors.
  • The company is led by co-founder and CEO Kevin Owens and is headquartered in Washougal, Washington.
  • Resect AI says its technology examines AI model behavior in real time to identify and intervene before hallucinated information reaches users.
  • The funding will support research and development, go-to-market initiatives, and hiring in the Seattle and Portland markets.
  • The company has not publicly identified its investors, disclosed a valuation, or classified the financing as a seed or Series A round.

Resect AI Emerges From Stealth With $25 Million

Resect AI launched publicly on September 3, 2026, with $25 million in funding from private-equity investors and a focus on one of enterprise artificial intelligence’s persistent problems: unreliable or fabricated model outputs.

The startup is developing what it calls an “accountability layer” for AI systems. Rather than building another general-purpose AI model, Resect AI is working on technology intended to examine how existing models behave and provide organizations with more visibility into their decisions.

The company did not disclose the identities of its investors, a valuation, or a conventional funding-stage designation such as seed or Series A. Its announcement said the capital will be directed toward research and development, go-to-market initiatives, and talent acquisition in the greater Seattle and Portland markets.

Resect AI is headquartered in Washougal, Washington, across the Columbia River from the Portland metropolitan area. Independent reporting from GeekWire said the company also plans to establish a Seattle-area office to support engineering and business operations.

AI Hallucinations Define the Company’s Technical Focus

Resect AI is targeting hallucinations, the inaccurate or fabricated responses that generative AI systems can sometimes present as factual information.

The company’s central technical claim is that its technology works “in-stream,” examining activity inside large language models as they operate rather than checking only the text generated after a response is completed.

According to Resect AI, the system is designed to observe, detect, interpret, and modify model behavior before a hallucinated response reaches the user. The company also says the technology can generate an audit trail for enterprise governance and compliance.

Those capabilities are important to distinguish as company claims. Resect AI has described how its technology is intended to operate, but the launch announcement does not provide independent testing data establishing how reliably the system identifies or prevents hallucinations across different models and applications.

That distinction matters as companies reconsider their enterprise AI software strategies and move generative AI from experiments into workflows where inaccurate outputs can carry operational or financial consequences.

Kevin Owens Frames Accountability as an Enterprise Issue

Co-founder and CEO Kevin Owens is leading Resect AI alongside co-founders Tim Walton, Tyler Gerber, and Tommy Lofgren.

Owens has positioned the company’s work around the level of trust businesses place in AI systems.

“AI has prematurely been put in a position of trust,” Owens said in the company’s launch announcement.

Resect AI says its technology is intended for organizations that require greater transparency into how AI models arrive at outputs, particularly when the systems are used for tasks where factual accuracy and oversight are important.

That approach places the company within a growing enterprise technology market focused not simply on deploying AI, but on governance, observability, security, and control around those systems.

The accountability question has become more prominent as businesses invest heavily in AI infrastructure and cloud capacity, alongside broader AWS and AI growth across the enterprise technology market.

The Accountability Layer Goes Beyond Post-Output Monitoring

Many AI monitoring systems evaluate generated responses after a model has already produced them. Resect AI is attempting to differentiate its approach by intervening earlier in the process.

The company describes its technology as operating inside the model’s processing stream, where it is intended to identify behavior associated with hallucinations before an unreliable answer is completed.

If the approach performs as described, the distinction could give organizations another way to oversee models used in production systems. Resect AI is also developing audit capabilities intended to record model activity for governance and compliance purposes.

The product is therefore positioned less as a replacement for an AI model and more as an additional control layer between the model and the organization deploying it.

Resect AI has also made its interpretability technology available through an open-source product, while pursuing commercial applications around enterprise accountability. Its launch places the company among a broader group of startups developing infrastructure and control systems around large language models rather than competing directly to build foundation models.

Enterprise Adoption Raises the Stakes for Reliable AI

Hallucinations are not a new limitation of generative AI, but their consequences can become more significant when models move into customer-facing or operational environments.

An inaccurate answer generated during an internal experiment may have limited consequences. The same problem can carry greater risk when AI is connected to business processes, customer communications, research, financial analysis, or other applications where users expect reliable information.

Resect AI is building its business around that transition.

Its stated objective is to give enterprises more insight into model behavior while providing mechanisms designed to intervene when a system begins producing unreliable information.

The company has identified publishing, finance, healthcare, research, and education among the industries where it believes greater AI accountability could be relevant. These are markets named by Resect AI rather than disclosed customer relationships.

The company has not announced major enterprise customers as part of its launch, so its next challenge will be demonstrating how its technology performs in production settings and whether organizations adopt an additional accountability layer alongside existing AI infrastructure.

Funding Supports Research, Commercialization and Hiring

Resect AI plans to divide its $25 million in funding among technical development, commercialization, and recruitment.

Resect AI Launches With $25M to Tackle AI Hallucinations

Photo Credit: Unsplash.com

Research and development spending will support continued work on its AI accountability and interpretability technology. Go-to-market investment will focus on bringing those capabilities to enterprise customers as the company moves beyond stealth.

Hiring is another immediate priority. GeekWire reported that Resect AI had about 30 employees at the time of its launch and planned to increase its workforce to 50 by the end of 2026. The company’s recruitment efforts are expected to include the Seattle and Portland markets.

Its Washougal headquarters gives the startup an unusual geographic profile for an AI company raising significant private capital. The company has said its presence in the Pacific Northwest is partly tied to access to technical talent across Washington and Oregon.

The funding gives Resect AI additional resources to develop its technology, but the company’s longer-term position will depend on whether its approach can demonstrate measurable reliability and enterprise value beyond the claims made at launch.

Resect AI Enters a Growing AI Governance Market

Resect AI’s launch comes as enterprises increasingly face questions about how to supervise generative AI systems once they are deployed beyond controlled experiments.

For organizations, the issue extends beyond whether an AI system can generate useful responses. Businesses also need ways to understand when a model is uncertain, document its behavior, manage inappropriate outputs, and determine who is responsible when automated systems make mistakes.

Resect AI is targeting that layer of the technology stack.

Its strategy centers on providing greater visibility into model behavior and intervening before unreliable information reaches users. The company’s $25 million financing gives it capital to pursue that approach through continued research, hiring, and commercial development.

The launch establishes Resect AI as a new entrant in AI accountability technology. The next test will be whether its in-stream approach can deliver the accuracy, transparency, and control that enterprises require when deploying AI in higher-stakes environments.

Frequently Asked Questions

What is Resect AI?

Resect AI is a Washougal, Washington-based startup developing AI accountability technology for enterprises. The company says its system is designed to examine model behavior in real time, detect potential hallucinations, and provide organizations with greater oversight.

How much funding did Resect AI raise?

Resect AI emerged from stealth on September 3, 2026, with $25 million from private-equity investors. The company plans to use the money for research and development, go-to-market activities, and hiring.

Who founded Resect AI?

Kevin Owens is co-founder and CEO of Resect AI. The other co-founders are Tim Walton, Tyler Gerber, and Tommy Lofgren, according to reporting published when the company emerged from stealth.

How does Resect AI address AI hallucinations?

Resect AI says its technology examines activity inside AI models in real time rather than relying only on checks performed after text has been generated. The company says this approach allows it to detect, interpret, and modify problematic model behavior, although those performance claims have not been independently established in the company’s launch materials.

Where is Resect AI based?

Resect AI is headquartered in Washougal, Washington, near the Portland metropolitan area. The company is also planning a Seattle-area office to support engineering and business operations.