Your staff are already using artificial intelligence. Your clients certainly are. So, how are you using this to your advantage? Will benign neglect leave you at risk? Are you leading how it changes your organization and your community?
AI has moved from experiment to everyday practice. Staff can readily use available tools to draft reports, summarize documents, analyze data, prepare presentations, research prospects, and personalize communications. They can often do it without your permission or without formal organizational recognition.
That should get the attention of every economic development organization leader. Yes, it offers massive possibilities for productivity improvements. It also offers frightening opportunities for mishaps that will permanently damage your street cred.
As CREC argues in our new white paper, Harnessing Artificial Intelligence to Transform the Economic Development Ecosystem, AI can greatly extend what an EDO can accomplish. It can also expose confidential information, produce convincing but inaccurate analysis, create contractual problems, and undermine the credibility that economic development organizations have spent years building.
The leadership challenge is not to stop AI. It is to use it well.
What’s Your Strategy for Adoption?
AI adoption isn’t really a technology decision. Which platform should we buy? Who needs a license? Should staff take a prompt-writing course?
Those questions matter, but they miss the larger opportunity.
The organizations that benefit most from AI will redesign how they work. Buying an AI tool and asking it to perform yesterday’s tasks a little faster will produce some efficiencies. However, the larger gains come when organizations ask a different question:
What could we now do better, faster, or at a scale that was previously impossible?
An EDO might personalize business retention outreach across hundreds of companies. It might analyze complex labor market information faster. It might conduct deeper due diligence on potential projects. It might automate routine administrative work so staff spend more time with employers and partners.
The advantage comes from redesigning the work, not merely adding another tool.
And that makes human expertise more valuable to deciding on your strategy.
AI works best when an experienced practitioner knows what question to ask, recognizes when an answer does not make sense, and understands the economic and political context behind the data. Domain knowledge and professional judgment become complements to AI.
Your credibility is your greatest AI asset
We all know that economic development is a relationship business. So, when every organization can generate a polished prospect letter, industry profile, or market analysis in minutes, polished content becomes less distinctive. Trust becomes more distinctive.
That creates an important paradox. AI can help us automate more communication, but the result should be more time for human relationships, not less.
The same principle applies to analysis. AI can help collect, clean, summarize, and interpret vast quantities of information. But an EDO still has to stand behind the final product. An hallucinated wage number in a board report or an invented citation in a site-selection response can erase years of hard-earned credibility.
AI should help us produce stronger work, not faster errors.
That means credible AI practitioners will do five things consistently:
Govern their own AI use. Decide which tools staff can use, what information they can enter, and what requires human review.
Assign responsibility. Give someone clear ownership for keeping guidance on AI use current.
Show their work. Verify data, sources, and conclusions before information leaves the organization.
Build judgment alongside access. Continue to train staff in the underlying economic development craft so that the technology is more meaningful.
Lead from experience. Use AI on real economic development work before advising businesses how to use it.
Then turn outward
As so many studies are now telling us, the bigger economic development opportunity does not sit inside the EDO. It sits across the regional economy.
Large companies have staff, consultants, and capital to test new technologies. Small and midsize businesses often do not. AI adoption therefore risks becoming another source of competitive separation between companies and between regions.
Economic development organizations can help close that gap, but only if they are prepared and build their own expertise.
EDOs already sit at the intersection of employers, workforce organizations, educational institutions, government, and community partners. That makes them natural brokers for regional AI adoption.
They can help smaller companies find trusted technical assistance. They can work with workforce and education partners to build AI-ready talent. They can identify the industries where AI adoption could produce the greatest productivity gains. And they can convene regional coalitions around a shared competitiveness strategy.
But there is an important prerequisite.
We cannot credibly ask the businesses in our regions to transform themselves with AI if our own organizations have not learned how to use it responsibly.
So what do we do? Start small. Start now.
An EDO does not need a major technology initiative to begin. Pick one or two pieces of real work that your organization performs repeatedly. Perhaps it is a labor market profile, an incentive review, business outreach, board research, or a program evaluation.
Run the task with AI for several weeks. Put an experienced staff member in charge. Verify every output. Measure the time saved. Track errors. Compare the result with your normal process. Document where the tool helped and where human judgment mattered most.
That small experiment can do three things. It builds staff capability and confidence. It creates evidence about where AI actually adds value. And it exposes the practical questions that your organization’s AI guidance needs to answer.
Then use what you learn to help your region, including your companies, your political leaders, and your partners, to become more AI savvy.
Economic development organizations have always helped communities respond to technological change. AI presents that challenge again, but this time it also changes the organizations doing the helping.
The regions that gain the most will not simply have the most AI tools. They will have businesses, workers, institutions, and economic development leaders who learn how to use those tools wisely.
The opportunity for EDOs is bigger than adopting AI. It is becoming the trusted institution that helps an entire region adopt it well.
Data centers are the most visible physical form of the AI economy. They are arriving faster than many states and communities can evaluate them. They bring billions in private investment, expand local tax bases, and fund grid, fiber, and water improvements. But they also raise questions about electricity demand, water use, air quality, and whether communities get fair value for the incentives they grant.
This white paper offers state, regional, and local leaders a grounded, evenhanded account of both sides. It argues that data centers are best understood as enabling infrastructure—like ports, airports, and electric grids—whose value lies in the economic activity they support rather than the handful of permanent jobs they create. Drawing on new causal research, case studies from Loudoun County to Lenoir to Lancaster, and the wave of legislation now moving through more than 30 states, the paper shows why benefits concentrate where deals are well structured and thin where communities accept investment figures at face value.
With data center development shifting toward rural areas that often have the least capacity to assess the tradeoffs, the paper lays out the questions leaders should be asking, principles for responsible decision-making, and the distinct roles state, regional, and local leaders each play. Above all, it makes the case that public trust and sound governance are themselves competitive assets in the age of AI.
The United States needs a common, trusted language for skills that is timely, local, and useful for real decisions. O*NET could provide the public foundation, but to fulfill that role, it must modernize through local detail, open standards, validation, and governance that responds to users.
Labor markets are local, but much of our data is not
Technology changes work first inside firms. Employers adopt new tools, reorganize tasks, and redefine jobs before those changes appear in national statistics.
These differences matter to state data offices, workforce boards, economic development organizations, colleges, and regional partners. They help employers find workers, guide students toward careers, and decide which training programs merit investment. When national data lack detail or currency, these organizations often build their own analyses. That can solve an immediate problem, but it also creates duplication and makes comparisons harder.
O*NET should provide the common language
O*NET represents a valuable inventory of knowledge, skills, and abilities, but it should not try to replace job postings, wage records, employer input, education data, or private analytics. Instead, O*NET should serve as the neutral public reference that allows these sources to connect. It can provide common definitions, identifiers, occupational descriptions, crosswalks, and validation standards.
Surveys continue to be invaluable as national benchmarks. Job postings can reveal changing demand. Wage and education records can show which skills and credentials lead to employment and earnings. Artificial intelligence can help organize text and identify emerging skills. A modernized O*NET can help provide a shared structure for the myriad sources of skills data.
Trust must remain the foundation
Faster data is not always better data. Job postings may reflect employer wish lists. Private taxonomies may rely on unclear methods. AI-generated classifications may reproduce bias or imply more precision than the evidence supports.
Skills data should show where it came from, how it was produced, what its limits are, and whether it has been tested against outcomes. O*NET should remain open to challenge. It should earn adoption by proving that it supports better decisions, not through mandate.
Private and alternative taxonomies should support comparison, validation, and possible inclusion. Independent evaluation and community engagement can help improvements flow back into the public standard.
Eight actions for a modernized system
The paper recommends that federal and state partners:
Modernize O*NET with validated AI and local detail.
Create interoperable standards.
Build state capacity.
Fund state experimentation.
Require third-party evaluation.
Create public-private governance.
Strengthen legal and data-sharing systems.
Provide durable funding for the public backbone.
The goal for O*NET is not one national database of skills. It is a common reference that helps employers, workers, educators, and public leaders make better decisions.
Arlington, VA – The Center for Regional Economic Competitiveness (CREC) is pleased to announce the election of Dominique Halaby as Vice Chair and the appointment of Jill Foys and Adrienne Johnston to the CREC Board of Directors. Their terms began on July 1, 2026.
Dominique Halaby serves as Associate Vice President for Innovation at Georgia Southern University, where he oversees the Business Innovation Group and the university’s innovation and technology transfer initiatives. Previously, he directed the Center for Community and Business Research at the University of Texas at San Antonio and led workforce and economic development initiatives in the Rio Grande Valley. He is Past President of the University Economic Development Association and serves on several boards and commissions.
Jill Foys serves as Executive Director of the Northwest Pennsylvania Regional Planning and Development Commission, where she leads regional economic and community development initiatives across eight counties. Throughout her career, she has focused on regional economic development, workforce partnerships, and strengthening collaboration between education and industry. Foys previously served as President of the Local Development District Association of Pennsylvania.
Adrienne Johnston is President and CEO of CareerSource Florida, where she leads Florida’s statewide workforce development system and oversees strategic workforce policy and investment initiatives. Previously, she served as Deputy Secretary of Workforce Services and Chief Economist for Florida Commerce, leading data-driven workforce and economic research initiatives that support businesses, workers, and policymakers across the state.
“Dominique, Jill, and Adrienne bring outstanding expertise and leadership to the Board,” said Ken Poole, CEO of CREC. “Their experience will help advance CREC’s mission and strengthen our impact.”
Founded in 2000, CREC is a nonprofit organization dedicated to helping regions compete and prosper by providing research, technical assistance, data analytics, and strategic guidance. Working with public, private, nonprofit, and academic partners, CREC develops innovative solutions that strengthen workforce systems, support economic development, and improve decision-making through high-quality data and analysis.
Media Contact: Ken Poole President & CEO CREC kpoole@crec.net – 703-504-2866
What a $151 billion demand signal teaches economic and workforce developers about certainty
Economic and workforce developers live by an unforgiving rule: before a region can train anyone, someone must say which jobs, how many, where, and when. The sharpest current test of that rule comes from an unexpected direction, the federal government’s new homeland missile defense initiative and the contract vehicle behind it. The story is set in defense, but the lesson travels to any region, any industry, and any training dollar that has to be spent before the future announces itself.
On January 15, the Missile Defense Agency completed the last of three award tranches under SHIELD, the Scalable Homeland Innovative Enterprise Layered Defense contract, the $151 billion vehicle that will carry most of the Trump Administration’s Golden Dome missile defense initiative. More than 2,400 companies now hold a position on it, and across the missile defense industrial base the hiring has already begun. The workforce development system cannot answer this demand signal, and the reason is not the one usually given. The problem is not only that the workers are missing. The problem is that the demand has no shape, and training systems can only aim at shapes.
The Shortage Everyone Can See
The familiar half of the story is well documented. The most recent Aerospace Industries Association (AIA) and McKinsey (2025) workforce study counts 2.21 million workers in the American aerospace and defense sector and reports attrition holding near 15 percent, roughly double the average across other United States industries. Seventy-six percent of member firms report sustained difficulty hiring engineering talent; 56 percent report the same for the skilled trades. Against the production surge now underway, an analysis published by Voyager Technologies (2026) puts the sector’s skilled technical deficit above 200,000 workers and estimates that the labor pool serving programs like Golden Dome would need to grow 30 to 40 percent to meet demand.[1] These figures are contested at the margins, but no serious observer disputes the direction. If Golden Dome were an ordinary program, this would be an ordinary, if large, shortage story.
The Structure Few Are Reading
Golden Dome is not an ordinary program, and the difference sits in the acquisition mechanics. SHIELD is a ten-year, indefinite-delivery, indefinite-quantity contract. The $151 billion is a ceiling, not an appropriation; the agency obligated no funds at the base award, and revenue flows to a vendor only when it wins a task order. The awards themselves went out in three tranches over six weeks: 1,014 companies on December 2, another 1,086 on December 18, and a final 340 on January 15 (Defense One, 2025; Defense Security Monitor, 2026). The money actually in hand is real but far smaller: roughly $24.4 billion provided through the 2025 reconciliation act and $13.4 billion in the fiscal year 2026 defense appropriation, with the administration planning to seek some $17 billion more through reconciliation next year (Federal News Network, 2026a, 2026b). Congressional appropriators, meanwhile, complain that they have received no master deployment schedule, no cost schedule, and no finalized system architecture (Federal News Network, 2026a; see also Defense One, 2026). If the members who fund the program cannot learn where and when the work will occur, a community college dean has no chance.
What Training Systems Need to Aim At
This matters because every workforce program that has ever delivered at scale was built against knowable demand. The scholarship names the same failure mode: MIT’s Task Force on the Work of the Future concluded that American training institutions are capable but fragmented, and that workers and firms alike underinvest in skills when the return is uncertain (Autor et al., 2022).[2] Consider the counterexample from submarine country. The Navy awarded Electric Boat a $15.38 billion contract modification in March (Reed, 2026); the yard committed to 8,000 hires this year (InsideDefense, 2026); and the New England training partnership that feeds it passed 10,000 workers trained on July 1 (turnto10, 2026; What’s Up Newp, 2026).[3] That pipeline works because five things are known: the employer, the occupations, the counts, the location, and the dates. Golden Dome inverts every term. The employer is any of 2,400 firms. The occupational mix depends on which architecture survives, and there is no finalized architecture. The locations follow task orders that have not been issued. The dates are whatever the ceiling becomes when it turns into orders. A regional consortium can train welders against a submarine contract. No one can train anyone against a ceiling.
What Firms Do in the Meantime
In the absence of shape, firms that can afford to pre-position are doing so with their own capital. True Anomaly raised $650 million in April, plans to nearly double its workforce to 500 employees by the end of the year, and intends to grow its factory footprint from 140,000 square feet toward two million over four years (CNBC, 2026).[4] Northrop Grumman has pointed part of a $1.65 billion capital expenditure budget at readying its Space Park campus for rapid production cycles (Voyager Technologies, 2026). These are rational hedges, and they have a predictable labor market consequence: firms that cannot see demand hire ahead of it, and firms that cannot hire ahead of it recruit from one another. With sector attrition already near 15 percent, much of the apparent hiring boom is the same cleared, experienced workers cycling among badge colors at rising wages. The deficit does not close, it circulates.
The Regional Gamble
The same uncertainty falls hardest on regions. Golden Dome work will eventually concentrate somewhere; the obvious candidates, given agency and prime contractor footprints, include Huntsville, Colorado Springs, Southern California, and Florida’s Space Coast, and firms like True Anomaly are already placing their factory bets in Colorado. But no governor can know today whether their state is on that list, and the list itself will be written one task order at a time. A steady-state shortage lets a region invest where its employers already are, the way southern Arkansas built training around a munitions cluster that has expanded sharply in the past three years, with Lockheed Martin, General Dynamics, and the Rafael-Raytheon joint venture R2S all adding capacity on a defense footprint dating to the 1940s (Arkansas Money & Politics, n.d.; RTX, 2025).[5] A shapeless shock forces the opposite calculation: gamble scarce training dollars on demand that may materialize three states away, or wait for certainty and forfeit the first-mover position that regional economic developers spend careers pursuing. Either choice is defensible. Both are expensive. And the cost is borne by exactly the institutions, community colleges and state workforce agencies, least able to absorb a wrong bet.
The Case for the Ceiling
The acquisition logic behind SHIELD deserves its due. A flexible, broadly competed vehicle lets the government move at commercial speed, avoids betting the program on a single architecture that intelligence or technology may overturn, and keeps two thousand innovators in the game rather than three primes. Judged purely as procurement reform, it may prove to be good design; that question belongs to others. The point here is narrower. Flexibility for the buyer is uncertainty for everyone downstream, and uncertainty does not disappear when it is exported. It comes to rest on the smallest actors in the system: the supplier deciding whether to add a second shift, the community college deciding whether to stand up a program, the technician deciding whether the job will exist in three years. The Harvard Project on Workforce has mapped that system as it actually exists: thousands of mostly small, local training providers, exactly the institutions least equipped to absorb the risk that flexibility exports (Project on Workforce at Harvard, 2023). The federal government has spent decades asking training systems to be demand-driven. Golden Dome is what happens when the demand declines to specify itself.
What Mobilization Has Always Required
There is precedent for both paths. When community colleges and certification bodies aligned curricula and credentials to the Artemis and Commercial Crew programs, from the Louisiana college that mapped its coursework to the technician tasks at the plant building the Artemis core stage to the FAA-recognized certifications anchoring Space Coast apprenticeships, the alignment worked because there were named vehicles, named sites, and named dates to align to (NASA, n.d.; Nunez Community College, n.d.; SpaceTEC, n.d.).[6] The Strategic Defense Initiative of the 1980s offers the other lesson: a loud national demand signal that never resolved into stable production, and left behind little durable workforce infrastructure. Which precedent Golden Dome follows will be decided less by the appropriated billions than by the task orders, because task orders are where a ceiling acquires an address, an occupation, and a start date. The industrial base has been asked to mobilize, and it is trying. But mobilization has never been a function of enthusiasm or even of money. It has always required someone to say what, where, and when. Until the task orders say so, the 200,000-worker question is not hard to answer. It is impossible to ask precisely. You cannot train against a ceiling. You can only train against an order.
Footnotes
[1] Voyager Technologies is a defense technology firm with a commercial interest in Golden Dome, and its figures characterize Aerospace Industries Association and McKinsey data. The underlying study reports a 2.21-million-person workforce, attrition near 15 percent, and hiring difficulty rates of 76 percent for engineering and 56 percent for the skilled trades; it does not itself state a 200,000-worker deficit. Treat these estimates as the analyst’s synthesis, not a study finding.
[2] The MIT Task Force on the Work of the Future reported in 2020; the book is cited here for its durable institutional findings, not for current labor market data.
[3] The partnership, run by the Southeastern New England Defense Industry Alliance (SENEDIA), holds a $98.3 million contract to train roughly 8,600 more workers through 2029.
[4] Company-announced targets, not audited filings or realized headcount.
[5] The R2S joint venture opened its East Camden plant in November 2025 and holds a $1.25 billion Tamir production contract (RTX, 2025); General Dynamics and the Army opened a $110 million load, assemble, and pack facility in April 2025. See also the companies’ pages: Lockheed Martin (n.d.), General Dynamics Ordnance and Tactical Systems (n.d.), and Rafael Advanced Defense Systems (n.d.).
[6] The Space Coast credentials are anchored by SpaceTEC, host of the Space Coast Consortium Apprenticeship Program (n.d.). The alignment was built by the colleges and certification bodies against the programs’ known requirements, not directed by NASA.
The race to build data centers is reshaping local economies faster than almost any other real estate trend. From 2023 to 2024, data center construction spending increased by 70%, more than triple the rate of any other property type. [i] The sector is drawing billions in venture capital and now consumes up to eight percent of total U.S. electricity.[ii]
For local communities, that growth is arriving with both a windfall and a set of tradeoffs: improved fiber and broadband networks, short-term and permanent job gains, and large tax receipts on one side, with infrastructure demands, public health concerns, and general local opposition on the other. Data centers are coming; the question is how economic developers can help communities capture the benefits and manage the tradeoffs.
Why are there so many new data centers?
A data center is a facility used to house large computer systems.[iii] There are many types of data centers hosting different services, but today’s headline centers are mostly used for cloud computing, including massive hyper scalers, which host at least 5,000 servers and can be 60,000 square feet in size.[iv] Cloud computing is an on-demand computing service, which can be rented by consumers. These centers are used for various IT purposes like storage, networking, and software deployment.
Separately, AI training data centers use cloud-computing’s on-demand infrastructure with specialized hardware and storage to train Large Language Models (LLMs), like ChatGPT. The AI market is booming and driving data center growth through investments by such companies as Microsoft, OpenAI, Google, and AWS. As these companies search for sites, EDO leaders are often the ones asked to explain what a data center will mean for a community. This piece is built for those conversations: a guide to the benefits, costs, and tradeoffs that come up when a project is proposed.
What are the potential benefits?
New tax revenues are the most significant benefit of hosting a data center. Depending on state and local tax regimes, new centers may pay real property taxes, personal property taxes, business income taxes, and sales and use taxes. Revenue can particularly surge in states which impose a personal property tax.
Some communities also benefit from developer philanthropy. Though unlike tax receipts, this giving is rarely guaranteed. A good example is Meta’s Data Center Community Action Grants program, which funds local STEM education and community projects.[v] A Meta facility in Huntsville, Alabama contributed to broadband expansions that brought internet access to nearby rural communities.[vi] While some communities benefit from data center philanthropy, , most charitable giving is contingent on specific Public Benefit or Community Benefit Agreement (PBA, CBA). These agreements typically stipulate investments in local workforce, agreements to limit the use of resources like water and energy, and investment in local funds. One such agreement in Cedar Rapids, Iowa required data center developers to invest in the local Community Betterment Fund, administered by the city council.[vii]
While data centers offer job gains, the highest number are related to the construction stage with permanent jobs, while high-paying, typically number less than 100 per site. Moreover, the scale depends on facility type and may take years to materialize. A Brookings analysis found that counties receiving their first large data center saw total private employment grow by 2,000 to 4,000 jobs over six years. Construction and IT sectors see the sharpest gains, but the overall employment effect is modest relative to the scale of investment these facilities represent.[viii]
What are the drawbacks?
Perhaps the most urgent challenge posed by data centers is energy use. A single modern AI data center can use energy equivalent to 100,000 homes over a year, and even larger data centers are entering the pipeline further increasing the demand.[ix] This creates a major challenge for utilities, community leaders, and economic developers, especially if this power demand results in higher costs for other consumers.
Data centers can also affect local air quality. To avoid using congested municipal power grids, some data centers utilize on-site natural gas turbines for power. These turbines – as well as diesel backup generators[x] at grid-connected centers – produce fine particulate matter and nitrogen oxides.[xi] The U.S. Environmental Protection Agency identifies both[xii], [xiii] as causes of harm to human health, but it remains difficult to isolate how much of these pollutants any single facility adds to a community’s air.[xiv]
As with energy, data centers need large amounts of water to cool their servers. In some cases, the water is taken from municipal sources which can raise water prices for residents. However, water consumption depends heavily on the type of cooling a data center uses. Centers that use closed-loop or immersion cooling consume less than those using evaporative cooling.[xv] Large data centers, like those that train LLMs for AI, can use as much as 5 million gallons per day.[xvi]
Taken together, the energy, air quality, and water demands of data centers represent costs that often fall on residents rather than developers. That doesn’t mean communities should turn data centers away, but it does mean the terms of any agreement matter enormously.
Are data centers right for your community?
Like every expansion project, communities should treat every transaction as a negotiation ensuring that any tradeoffs are more than offset by project benefits. Because demand for sites is high, developers need communities as much as communities need them, and that gives community leaders room to negotiate. Some Community Benefit Agreements have required developers to invest in local workforce programs, infrastructure, and community funds. Others have secured commitments to reduce grid electricity demand and water consumption.
What undermines this leverage is the incentive package. Many communities are offering tax exemptions and other concessions to attract developers, which can minimize potential tax revenue that makes data centers attractive. This is a particular risk given the relatively low job creation for residents.[xvii]
The financial case for data centers rests heavily on their tax contributions, and those contributions can erode quickly when lofty incentives are on the table.
Major Takeaways
Data center development is moving fast, and the communities best positioned to benefit are those that have assessed their leverage, tax structure, and infrastructure capacity before entering negotiations.
Tax revenue can be substantial, but incentives offered to attract developers can offset those gains
Data centers are not major job creators and generate modest local wage increases, so financial benefits largely depend on tax structure
Energy and water demands can raise costs and health risks for residents, particularly in densely populated areas
Community Benefit Agreements that require developers to invest in the community including workforce programs and infrastructure
The type of data center matters: cooling systems, on-site power generation, and facility size all affect local impact
Data centers are less about job creation and more about tax base expansion, infrastructure investment, and long-term economic positioning. For communities, the real value comes when leaders align these projects with broader development goals while carefully managing tradeoffs like energy demand, land use, and community concerns. EDO leaders can use this information to help communities enter negotiations with a clear picture of what a given project will actually cost and deliver for their community before extending incentives.
Artificial intelligence (AI) is being adopted faster than any technology in modern history. Within two years of widespread availability, 39.4 percent of American adults were using generative AI tools, a pace that outstripped both the personal computer and the Internet at comparable stages (Deming, 2024). Yet the defense industrial base (DIB), the network of manufacturers that produces everything from fighter jet components to submarine hull plates, has no structured workforce training program to match this adoption curve. The technology is arriving. The people who need to use it are not ready.
That disconnect is the central finding of a 32-source analysis conducted by the Center for Regional Economic Competitiveness (CREC) under a cooperative agreement with the Department of Defense (DoD) Manufacturing Technology (ManTech) Program Office. The research draws on peer-reviewed studies, policy documents, industry analyses, and workforce data to answer a straightforward question: what does AI actually mean for the defense industrial base over the next five years? The answer is more nuanced, and more urgent, than most of the conversation around AI in defense would suggest.
The Workforce Problem Is the AI Problem
The bottleneck in defense manufacturing is not access to AI. It is the availability of workers who can use it. Manufacturing labor productivity has been nearly flat since the mid-2000s despite decades of automation investment. The missing ingredient is not more technology but more skilled humans: process engineers, electricians, robotics specialists, maintenance technicians, and quality inspectors. Deloitte projects 3.8 million net new manufacturing jobs by 2033. The question is not whether humans are needed but whether enough of them will be available with the right skills.
AI tools produce a median 25 percent productivity improvement when used effectively (Sadun, 2025), but that word effectively carries enormous weight. Without structured training and integration support, adoption produces uneven results and worker frustration rather than sustained gains. The Computer Numerical Control (CNC) revolution of the 1970s and 1980s provides the relevant precedent. Plants that adopted CNC were 75 percent more likely to use problem-solving teams, twice as likely to offer technical training, and held regular shop floor meetings at substantially higher rates than non-adopters (Deming, 2024). Automation changed the nature of skilled work but did not eliminate the need for it. AI will be no different.
AI Is Hollowing Out the Talent Pipeline
One of the most consequential findings in the research is a dynamic that almost no one in defense manufacturing is discussing: AI is absorbing the routine tasks that junior workers need to develop into senior experts. Anthropic’s own internal study found that while engineer productivity gains grew from approximately 20 percent to 50 percent year over year, senior engineers reported growing concern about skill atrophy among junior staff. Junior engineers stopped asking questions of mentors because AI answered faster (Orrell, 2025).
In defense manufacturing, this pattern is particularly dangerous. Senior machinists, welders, quality inspectors, and maintenance technicians acquire their expertise through years of hands-on practice that begins with routine work. If AI absorbs those tasks before workers have the chance to learn from them, organizations gain short-term productivity at the cost of a long-term senior talent shortage. The pipeline narrows even as current output rises.
The Technology Is Not Ready for What Defense Demands
Enthusiasm for AI in manufacturing often outpaces what the technology can reliably deliver. AI-powered Automated Optical Inspection (AOI) systems currently achieve 60 to 70 percent accuracy, a rate that is insufficient for defense applications where tolerances are measured in thousandths of an inch. On the agentic AI front, the best-performing architecture for manufacturing decision support, a Retrieval-Augmented Generation (RAG) system, achieved 77.89 percent accuracy compared to 52.37 percent for a baseline Large Language Model (LLM). Better, but not defense-grade.
The implication is straightforward: human-in-the-loop oversight will remain essential in defense manufacturing for the foreseeable future. Planning for AI in this sector means planning for augmentation, not autonomy. Workforce development strategies that assume AI will replace human judgment are building on a foundation that does not yet exist.
Implementation Matters More Than the Technology
The same AI technology can produce opposite outcomes depending on how it is deployed. Where AI is implemented primarily for surveillance and performance monitoring, workers report reduced autonomy, lower morale, and increased turnover. Where workers are involved early in design and deployment, 96 percent reported increased job satisfaction when freed from monotonous tasks to focus on higher-level work. Collaborative robots (cobots) can increase worker productivity by up to 85 percent when paired with proper training (International Federation of Robotics). For defense manufacturers, the management decision matters more than the technology decision. Procurement of AI systems without parallel investment in organizational change is a misallocation.
The System Is Not Built for This Speed
AI capabilities evolve on timelines measured in months. Defense acquisition spans years. Secretary of Defense Pete Hegseth has publicly stated that the current acquisition system is archaic and must shift from decade-long development cycles to rapid iteration. Small and mid-size defense suppliers cannot wait for multi-year contract modifications to adopt AI tools that may be obsolete by the time approval comes through.
Meanwhile, the data needed to make smarter workforce investments now exists. CREC and RTI International, under the ManTech cooperative agreement, produced the first national dataset measuring the Critical Technology Area (CTA) workforce in aerospace and defense manufacturing, delivered in December 2025. This dataset maps supply and demand for skills in AI, autonomous systems, additive manufacturing, and other priority technology areas across the top 30 aerospace manufacturers. The data confirms significant gaps between CTA skill supply and employer demand. Yet no federal workforce investment program currently uses it. The approximately $6 billion the federal government spends annually on workforce development through the Workforce Innovation and Opportunity Act (WIOA), the Perkins Act, and grants from the Economic Development Administration (EDA) and the National Institute of Standards and Technology (NIST) is allocated without reference to defense-specific workforce intelligence.
The Untapped Pipeline
Approximately 200,000 service members transition out of the military each year with technical training, leadership experience, security clearances, and familiarity with defense systems and culture. They represent the single most qualified talent pipeline available for defense manufacturing. Yet persistent mismatches between military occupational specialties and civilian job classifications, inadequate credentialing bridges, and employer unfamiliarity with military skill sets prevent this pipeline from flowing at scale. Community colleges, which should function as the primary bridge between military training and civilian manufacturing careers, vary widely in their alignment with local defense labor markets. The best performers, such as Dallas College, Wake Tech, and Miami Dade, maintain active employer relationships and anticipate skill demand. Most do not (Fuller, HBS).
AI is a necessary but insufficient condition for a competitive defense industrial base. The binding constraints are workforce readiness, organizational capacity to integrate new tools, acquisition pathways that match the pace of technology change, and analytical infrastructure that connects federal investment to measurable outcomes. Technology alone will not resolve any of them.
Deliberate action looks like this: structured AI training for the existing manufacturing workforce. Redesigned development pathways that ensure junior workers still build expertise even as AI absorbs routine tasks. Acquisition reform that matches technology timelines. And federal investment guided by actual workforce data rather than assumptions about where the gaps are. The data exists. The talent pipeline exists. The question is whether the institutions responsible for the defense industrial base will use them before the window closes.
Sources
Deming, D. (2024). “The Rapid Adoption of Generative AI.” NBER Working Paper.
Deming, D. (2024). “How Computers Turned Machinists Into Problem-Solvers.”
Fuller, J. (HBS). “Why the Skills Gap Persists.” Harvard Business School.
Linder, B. (2026). “AI Won’t Save Manufacturing.” Forbes.
Orrell, B. (2025). “What Anthropic’s Internal Study Suggests About the Future of Work.” American Enterprise Institute.
RTI International / CREC / ManTech (2025). “Measuring the Size and Dynamics of the CTA Workforce.”
Sadun, R. (2025). “Reskilling the Workforce With AI.” Harvard Business School.
“Agentic AI for Smart Manufacturing” (2025). SSRN.
“AI and Job Quality: Insights from Frontline Workers” (2024). Partnership on AI / SSRN.
Global events are reshaping the environment in which manufacturers operate, accelerating defense production and increasing pressure across supply chains. Workforce systems must be ready to respond.
Demand signals are sharpening. Defense suppliers are preparing for higher production levels, and many will need more technicians, machinists, quality specialists, and production workers. Intermediaries should be working with manufacturers now to identify surge occupations, map regional training capacity, and reduce the time between hiring demand and credential completion.
Training timelines must accelerate. Defense production schedules move faster than most traditional workforce programs. Manufacturers are increasingly looking for short-cycle training, modular credentials, and employer-driven curriculum updates that can move workers into production roles quickly.
Workforce visibility is becoming a competitive advantage. Manufacturers need real-time insight into workforce availability across regions and supply chains. Intermediaries that can provide labor market intelligence, talent pipeline data, and employer coordination will become essential partners in supply chain readiness.
Partnerships matter more than programs. No single organization can meet the scale of emerging workforce demand. Intermediaries that connect manufacturers with community colleges, training providers, workforce boards, and regional economic development partners will help firms expand capacity and reduce risk as production ramps up.
For workforce intermediaries, the coming months represent a window to act:
Work with manufacturers now to map surge occupations and talent gaps.
Align training programs with emerging technologies and defense production needs.
Strengthen partnerships across colleges, workforce boards, and industry associations.
Improve access to workforce data so manufacturers can better understand where talent is available.
At CREC, we are hearing increasing interest from regions and intermediaries looking to strengthen the workforce backbone of defense and national security supply chains.
If you work with manufacturers or workforce programs supporting advanced manufacturing, what are you seeing on the ground right now? Are firms signaling increased demand, supply chain pressures, or new workforce challenges? Where are the biggest talent bottlenecks emerging?
Let’s compare notes and talk about practical next steps. Connect with us.
On January 8, Congress released its proposed FY 2026 mini-bus appropriations for Commerce, Justice, Science budget. This budget reflects a recalibration for economic development, and it offers useful clues about how federal economic development policy is likely to shape state and regional practice in 2026 and beyond.
On the bright side in this environment, the budget provides a relatively flat $466 million funding level for EDA, only slightly less than FY 2024 and FY 2025. That alone is telling. Congress appears comfortable with the current scale of EDA. The real signal lies in the reallocation of dollars within that total.
Legacy line items like Technical Assistance and Trade Adjustment Assistance are receiving haircuts and the funds are being shifted to Economic Adjustment Assistance and Assistance to Coal Communities. While not dramatic, the cuts reinforce a longer-running shift away from generalized support functions toward more targeted, place-based execution. Furthermore, Recompete now appears as an $18 million line item. Congress clearly want to protect this effort to address the challenges in distressed areas that keep working age adults out of the workforce. Where it can, Congress is clearly prioritizing adjustment tied to real economic shocks and structural transition.
In addition, workforce development is clearly a rising EDA priority. Workforce Training Grants appear as a standalone $10 million line item for the first time. At first blush, one might see this as a rebranding of the soon-to-be defunct Good Jobs Challenge program. Likewise, the STEM Apprenticeship program continues at $2.5 million. Not enough funding for either program to create an impactful national workforce strategy, but these funding allocations send a signal that workforce development is a national economic development priority. Workforce has been embedded across EDA programs for years, but this year EDA is continuing to elevate industry-driven training as a priority. At some point, we may begin to see significant dissonance with this focus on human capital when the agency has traditionally emphasized Public Works and Planning activities focused on physical and community infrastructure investments.
There are also quieter signals worth noting. Biomass funding disappears as a distinct line item, indicating less appetite for narrowly scoped sector carve-outs within EDA. Salaries and Expenses decline by $2 million, reinforcing the commitment to streamline Federal agency staffing.
Taken together, these shifts reinforce a federal government not looking to expand the economic development toolkit. The investments place more emphasis on adjustment, workforce execution, and place-based competitiveness. We’ll also see more pressure to show results.
For economic development leaders, the takeaway is to better integrate workforce training, adjustment strategies, and industry engagement into a coherent execution model. Organizations that rely on planning grants or loosely connected initiatives will find the environment less forgiving. In that sense, the FY 2026 EDA budget is best read as a preview. It reflects how Congress expects economic development to function. Stable funding, but with fewer labels, more distinct priorities, and a higher bar for execution.
The start of a new year brings both momentum and responsibility to CREC and our partners to ensure that the nation’s defense industrial base is ready to respond to technological innovation. We must move beyond short-term responses to workforce shortages across all advanced manufacturing sectors and toward a more deliberate effort to build the technical capacity this sector will require over the next decade. The challenge is no longer simply filling positions; it is preparing workers to succeed in environments shaped by rapid technology cycles, digital engineering, advanced materials, automation, and increasingly compressed production timelines. Defense manufacturers are integrating AI-enabled systems, model-based design, additive manufacturing, and resilient supply-chain technologies; as such, the workforce must adapt as skill needs change
Employers, state leaders, and federal partners are seeking to modernize training models and create more accessible and affordable credentials to meet industry need. The approach to pedagogy is also changing to accommodate more immersive and simulation-based learning, replacing traditional classroom settings and supplementing work-based learning. Emerging pathways offer opportunities for reskilling and upskilling to better align with data-driven production, cybersecurity, robotics, and other critical technology areas.
From our perspective at CREC, we are especially pleased to see that more leaders are recognizing the importance of workforce development as a strategic enabler of defense readiness. The Department of Defense, through ManTech and its partner Manufacturing Innovation Institutes, are taking a more active role in translating the skills required to apply cutting-edge research in manufacturers by ensuring those firms have access to relevant training solutions, aligning those solutions to the delivery capabilities of education providers, and supporting efforts that help to build regional talent pipelines prepared to fulfill the advanced skill needs for both near-term production and long-term technological advantage.
We expect the pace of change in 2026 to accelerate. Technology-enabled manufacturing, cross-sector partnerships, and regionally grounded strategies are becoming the norm, so existing organizations will need to partner to adapt. At CREC, we remain committed to supporting the work of bringing key partners together and linking them with unique resources like the MIIs, but we are clear-eyed about what it will take to succeed. Lasting impact will depend on coordination across federal programs, sustained industry engagement, and continued investment in the institutions and intermediaries that connect people to advanced manufacturing careers. By strengthening these partnerships and fully leveraging ManTech’s EWD investments, we can ensure that America’s defense manufacturing workforce is ready for what comes next and be well positioned to lead it.
