October 2026

The first time a deadline slipped hard on my watch as a founder, I scanned the room for a name. Who dropped it? Who missed the handoff? Who should have flagged it sooner? The quiet that followed was the real signal. People were deciding how much of the truth was still safe to bring to me.

I remember that quiet more than the missed date.

Here’s my situation, so you know where I’m coming from. I run an SBA lender and certified community development financial institution, and I teach entrepreneurial leadership at Arizona State University’s W.P. Carey School of Business. Across those chairs I keep watching the same move: When something breaks, attention rushes to the person closest to the break. The conditions that made the break likely stay unnamed.

Most owners treat accountability as something that flows down the org chart. That direction is backward. Accountability is credible only when responsibility gets heavier as authority rises. A shop that disciplines frontline mistakes while insulating senior judgment is transferring risk, not demanding excellence. Gallup finds that managers account for at least 70% of the variance in team engagement, and only 20% of employees worldwide were engaged in 2025. The layer that shapes what people will say out loud is the layer that owns the conditions.

I used to treat the hunt for a name as leadership. Wrong move. Finding a person can be necessary. Starting there trains everyone else to edit the story before it reaches you.

We are living through a stretch of institutional stress that Neil Howe and others call a Fourth Turning: a period when trust in systems thins and character under pressure matters more than the playbook. In that climate, the posture I call Washington Mode is useful here: Restraining. Carry the power you hold without using it to shove risk onto people with less of it.

Own the conditions you designed first

After a miss, write the conditions before you write the name. What priorities did you set? What staffing did you approve? What incentives did you leave in place? What information never reached the person who had to act? Those are your levers. Leaders control more of the context than anyone else in the building, so they own a share of the failure environment even when they did not make the last move.

The principle fits in one line: Accountability is an information system before it is a verdict.

Try the same sequence this week. The next time something breaks, open with four questions you ask yourself out loud: What did I prioritize? Whom did I leave underpowered? What did we reward? What signals got ignored? A restaurant owner might find the late delivery was preceded by a schedule that never left room for receiving. A SaaS founder might find the outage followed a release rule everyone knew was theater. Name the conditions while the room is still listening.

Separate the person from the system, in that order

Upward accountability does not erase individual agency. Someone still made a choice. The test is sequence. First, ask what the system and the leader made possible, rewarded or ignored. Then judge the individual act. Reverse the order, and people learn that candor is expensive.

The principle: Own the environment first, then hold the person.

Say the split plainly in the room. “Here is what I set up that made this more likely. Here is the decision that still belongs to you.” A franchisee can do that after a bad weekend shift. A clinic owner can do it after a billing miss. A fabricator can do it after a scrap spike. The split keeps dignity intact and still draws a line. Skip it and you get either scapegoats or vagueness. Neither builds a team that can run without your mood as the weather system.

Make the next hard truth cheaper to deliver than to hide

The point of owning upward is not theater. It is to lower the personal cost of telling the truth the next time. When leaders absorb the share of failure they actually own, weak signals travel earlier. When they externalize blame, reporting becomes dangerous and the institution learns to hide. Trust compounds from evidence. A single clean ownership move after a miss is one of the densest deposits you can make.

The principle: The leader’s first job after failure is to make truth safe enough to surface.

Build the habit on a schedule, not only in crisis. After every meaningful miss, run a 10-minute debrief with the people who were closest to the work. You speak first about the conditions. They speak second about the choices. You close with one change to the system and one clear expectation for the person. Keep one annoyance unresolved on purpose if the room still feels too careful. Carefulness is data.

Here’s where I’d start this week: Pick the last miss that still bothers you. Write four lines on the conditions you created. Write one line on the choice that still belongs to someone else. Say both lines to that person before Friday. Then ask what they edited out of the story the first time. Watch whether the next hard truth arrives sooner.

The quiet in that early room was never really about a deadline. It was about whether the truth still had somewhere safe to land. Build that landing before the next miss arrives.

The first time a deadline slipped hard on my watch as a founder, I scanned the room for a name. Who dropped it? Who missed the handoff? Who should have flagged it sooner? The quiet that followed was the real signal. People were deciding how much of the truth was still safe to bring to me.

I remember that quiet more than the missed date.

Here’s my situation, so you know where I’m coming from. I run an SBA lender and certified community development financial institution, and I teach entrepreneurial leadership at Arizona State University’s W.P. Carey School of Business. Across those chairs I keep watching the same move: When something breaks, attention rushes to the person closest to the break. The conditions that made the break likely stay unnamed.

Early in my career, I remember sitting in a doctoral seminar, looking around the room and thinking, “I don’t belong here.” Everyone else seemed sharper, more prepared and more confident. Meanwhile, I was mentally editing every thought before I let it leave my mouth.

Eventually, I realized I had developed a pretty effective system for never embarrassing myself: Don’t say anything until I’m certain I’m right. Unfortunately, it was also an effective system for never growing. So I gave myself a different rule: If I’m not embarrassing myself at least once a day, I’m probably not pushing myself hard enough.

Decades later, after moving into entrepreneurship and venture investing, I’ve turned that lesson into a more disciplined approach. I call it Act, Adjust, Absorb.

Act: Know what ‘enough’ looks like

“Move before you’re ready” is terrible advice if it means ignoring evidence. Leaders need to distinguish between certainty, which is rare, and a sufficient signal, which they can actively look for.

When I began exploring what would eventually become Desert Forge Ventures, an early-stage venture capital fund focused on the Las Vegas ecosystem, I didn’t have every detail figured out. I felt limited in my ability to act, so I did what I’ve learned to do with consequential decisions: I talked to people who understood the opportunity, challenged my assumptions, and looked for evidence that the idea could work. I started by talking with my nephew, who is deeply involved in the Silicon Valley ecosystem. His advice helped me frame and focus the fund, preparing me for the wave of AI startups that was coming. I also spoke with a friend in Tucson who founded venture capital funds focused on the University of Arizona. His advice helped refine the fund’s positioning in Las Vegas. He also shared the legal documents from his first fund, giving me a practical template to work from.

These conversations were instrumental in shaping the fund. The key is not to seek validation for what you already believe, but to ask questions that might prove you wrong. What am I missing? What has to be true for this to work? Where could this fail? At some point, more analysis stops reducing risk and starts postponing action. That’s when leaders need to act.

Adjust: Make the first move small enough to learn from

My first two investments taught me the importance of starting small and remaining flexible.

I launched the fund with a strong interest in deep tech emerging from university labs, and our first two investments fit that thesis perfectly. Both companies, however, wanted fairly substantial lead investments, perhaps more than we were ready to make. I agreed to move forward even though we were still raising the necessary capital, and that put us in a difficult position.

To get out of it, I had to adjust. We accelerated fundraising, recalibrated conversations with investors and became much more deliberate about communicating timelines to both sides. That experience reinforced an important rule: Move early on reversible decisions, but proceed carefully when your decisions create obligations for others.

Today, before making a consequential commitment, I pressure-test it with five questions:

• What assumption am I testing?
• What is the smallest credible move I can make?
• What has to be true for this to work?
• Who becomes dependent on us if I say yes?
• What evidence would cause me to accelerate, change course, or stop?

Absorb: Turn experience into a better system

Leaders often act and adjust but skip the most valuable step: absorbing the lesson. We get busy and move on. But entrepreneurship gets easier when we treat this mindset like a muscle and strengthen it through small, repeated efforts. To put this into practice, set aside time each week to identify one conversation, question, or decision you’ve been avoiding. Address it early, then reflect on three questions:

What did I expect would happen? What actually happened? What did I learn?

My experience with the fund changed how I think about readiness. I no longer ask only, “Am I ready to move?” I also ask whether the people and resources affected by my decision are ready. I distinguish between what I know, what I believe, and what still needs testing. And I deliberately seek out people willing to challenge my thinking.

Experience should give leaders more than confidence. It should give them better filters.

Find your judgment zone

Entrepreneurial advice often encourages people to push beyond their comfort zones. The Act, Adjust, Absorb framework can help you do that, but when applied consistently, it offers something more valuable: a process for comparing the outcomes you fear with what actually happens. Over time, you build evidence. You learn that asking the supposedly stupid question often reveals that others were confused, too. You discover that respectful disagreement rarely leads to the catastrophe you imagined. And you begin to recognize which risks are real and which exist mostly in your head.

Earlier in my career, uncertainty sometimes felt like evidence that I wasn’t ready. Today, I see it as my judgment zone. I’m not fearless, but my decisions are better calibrated. I embrace uncertainty as an opportunity to learn, and while I’m certainly not immune to mistakes, I’m better at learning from them and applying those lessons to my next decision.

That’s the confidence gap leaders need to close. Don’t wait to feel confident before acting. Act when you have sufficient evidence, adjust when reality teaches you something, and absorb the lesson so your next decision is better. If you want more confidence as you contemplate your next move, take heart in Goethe’s words: “Doubt can only be removed by action.”

Early in my career, I remember sitting in a doctoral seminar, looking around the room and thinking, “I don’t belong here.” Everyone else seemed sharper, more prepared and more confident. Meanwhile, I was mentally editing every thought before I let it leave my mouth.

Eventually, I realized I had developed a pretty effective system for never embarrassing myself: Don’t say anything until I’m certain I’m right. Unfortunately, it was also an effective system for never growing. So I gave myself a different rule: If I’m not embarrassing myself at least once a day, I’m probably not pushing myself hard enough.

Decades later, after moving into entrepreneurship and venture investing, I’ve turned that lesson into a more disciplined approach. I call it Act, Adjust, Absorb.

Key Takeaways

• AI Overviews have transformed informational search, but traditional local-service searches remain heavily dominated by Google’s Local Pack and Maps results.
• According to Whitespark’s research into AI Overviews and local search, AI Overviews appeared on just 15% of traditional local-intent queries, compared with Local Packs appearing on 93%.
• For service businesses, ranking still matters because consumers searching for a nearby provider often need a business, not an AI-generated answer.
• Google Business Profiles may also provide useful information to the broader AI search ecosystem, making strong local SEO an investment in both traditional and AI visibility.

There’s a growing narrative in marketing that traditional search matters less every year. AI Overviews answer questions directly in Google, ChatGPT and Claude recommend businesses, and consumers discover companies on social media and other platforms.

All of that is true. But after a decade in local SEO, I’m still focused heavily on one decidedly traditional strategy: scaling Google Business Profiles and physical locations for local businesses.

The reason is simple: AI has disrupted informational search far more than the search for a local service provider. A person searching “roofing companies near me” or “personal injury lawyer St. Louis” isn’t looking for an explanation. They’re looking for a business. Google still has an extraordinarily effective product for that: Google Maps and the Local Pack.

Local search has resisted AI disruption

Whitespark’s 2026 Local Search Ranking Factors report describes local search as “insulated” from the AI disruption affecting the broader industry. Informational publishers have been hit hard, but local business discovery still happens mostly through Google’s local results.

A separate Whitespark study of AI Overviews in local-business searches shows why. It divided searches into three types of intent:

• Local queries, such as “personal injury lawyers in Phoenix”: AI Overviews appeared 15% of the time, while the Local Pack appeared 93% of the time.
• Informational queries: Whitespark found AI Overviews on 92% of searches and Local Packs on 6%.
• Hybrid queries: AI Overviews appeared 97% of the time, compared with 17% for Local Packs.

“How much does a new roof cost?” and “roofing company near me” may come from the same customer, but Google treats them very differently. That’s why local SEO still matters.

Ranking still matters when the customer needs a business

When someone needs a business, Google still shows the Local 3-Pack, and so do AI Overviews when they appear. That result is built from Google Business Profile data. It gives searchers nearly everything they need to decide, from reviews and hours to photos, services and directions, without a website click. That’s why so many local SEO professionals focus on maximizing Google Business Profile visibility. For single-location businesses, though, that’s usually where the strategy stops.

Location is a distribution strategy

For businesses that can scale, each new location creates another geographic point from which to compete. I’m not only asking how to get a website more leads, but how to expand the footprint from which a company acquires them.

A legitimate new location adds more than an office. It adds another Google Business Profile, another Maps presence, another review profile and another proximity point. For businesses that depend on local demand, a physical location is a customer-acquisition asset as well as an operational decision.

Google’s local algorithm is inherently geographic. Google’s own documentation names relevance, distance and prominence as the primary local ranking factors, and Whitespark’s 2026 Local Search Ranking Factors confirms that proximity remains a major signal for Local Pack and Maps visibility.

A highly authoritative website can still struggle in Maps searches far from its office. Opening legitimate locations closer to where customers search solves a problem that backlinks and content often can’t. For service businesses that cover large areas, like attorneys and home service companies, this approach has worked for years.

AI may make Google Business Profiles more important

AI systems still need real-world information: who you are, what you do and where you do it.

In January 2026, Local Falcon analyzed more than 50 million U.S. Google search results across 1,993 business categories. It observed that AI-generated local recommendations often drew on public information from Google Maps, reviews, ratings and Business Profiles.

That doesn’t mean a profile directly causes an AI recommendation; the relationship is still developing. But it reinforces a larger point. Accurate business data, service pages, strong reviews and third-party citations help wherever a customer finds you. The choice isn’t local SEO versus AI optimization.

Reviews are a conversion layer

According to BrightLocal’s 2026 Local Consumer Review Survey, 97% of consumers read online reviews for local businesses, and 71% use Google for those reviews, more than any other platform.

Reviews also decide whether visibility turns into a customer. Separate BrightLocal consumer-search research published in 2026 found that 65% of consumers who rejected a local business did so for at least one review-related reason, such as negative reviews, low ratings, or too few or outdated reviews.

A business can win the Local Pack impression and still lose the customer to a competitor with a stronger reputation one listing over.

Reviews also have a growing second audience: machines, which use their natural-language detail to understand what a business does.

Prepare for AI without abandoning what works

None of this means ignoring AI search. BrightLocal’s 2026 Local Consumer Review Survey found that use of ChatGPT and other generative AI tools for local recommendations jumped from 6% to 45% in a single year, making AI the third-most-popular source in the survey. Businesses should be thinking about mentions, citations, structured data and how AI represents their brands.

But preparing for where search is going is different from abandoning where customers are today. In March 2026, Local Falcon compared 189,905 ChatGPT results with millions of Google local results for restaurant queries. Local Falcon found that 83% of restaurants were completely invisible in ChatGPT, compared with only 14% in Google.

That study covers only restaurants, but it shows how far apart the two kinds of visibility remain. AI is a new discovery layer, not yet a replacement for the existing ones.

Local SEO is becoming more specialized

The biggest mistake marketers can make in 2026 is treating search as one channel. Informational search is changing fast, but transactional local search works differently because geography changes the answer.

Ask an AI system what compound interest is, and everyone gets essentially the same answer. Ask who should replace your roof, and location, availability, reputation and service area suddenly matter. Those answers require local signals, and Google has spent decades building an ecosystem to organize them. Local Services Ads also draw on Google Business Profile data, which adds another layer of protection.

The best strategy isn’t chasing the channel undergoing the biggest change. It’s finding where customers still show purchase intent and building a defensible presence there. For many local businesses in 2026, that place is still Google Maps. Until consumer behavior says otherwise, I wouldn’t bet against it.

Key Takeaways

• AI Overviews have transformed informational search, but traditional local-service searches remain heavily dominated by Google’s Local Pack and Maps results.
• According to Whitespark’s research into AI Overviews and local search, AI Overviews appeared on just 15% of traditional local-intent queries, compared with Local Packs appearing on 93%.
• For service businesses, ranking still matters because consumers searching for a nearby provider often need a business, not an AI-generated answer.
• Google Business Profiles may also provide useful information to the broader AI search ecosystem, making strong local SEO an investment in both traditional and AI visibility.

There’s a growing narrative in marketing that traditional search matters less every year. AI Overviews answer questions directly in Google, ChatGPT and Claude recommend businesses, and consumers discover companies on social media and other platforms.

All of that is true. But after a decade in local SEO, I’m still focused heavily on one decidedly traditional strategy: scaling Google Business Profiles and physical locations for local businesses.

The reason is simple: AI has disrupted informational search far more than the search for a local service provider. A person searching “roofing companies near me” or “personal injury lawyer St. Louis” isn’t looking for an explanation. They’re looking for a business. Google still has an extraordinarily effective product for that: Google Maps and the Local Pack.

A local AI box isn’t there to replace the cloud for large-scale training. It exists for three specific things: keeping sensitive data off someone else’s servers, avoiding metered API costs when you’re iterating hundreds of times a day and getting a fixed, predictable machine you fully control instead of a shared, rate-limited one. Cloud GPUs still win for large training runs and burst capacity. A local box wins for prototyping, fine-tuning smaller models and running inference on data you’re not allowed to send anywhere else. That’s the tradeoff, and it decides which of these three ecosystems actually fits your work.

NVIDIA, AMD and Apple are all shipping new local AI hardware on a near-constant cycle right now. Let’s look at which machine actually makes sense for you.

NVIDIA: the safe, expensive default

The NVIDIA DGX Spark pairs 128GB of unified memory with a Grace Blackwell GB10 chip built around CUDA, which is still the deepest, most battle tested software ecosystem in AI. Existing PyTorch code, TensorRT and cuDNN pipelines run on it without modification, which is not something you can say about the other two options yet.

The advantage is that nothing needs to be rewritten or reverse engineered. If your team already builds on CUDA, this is the lowest friction way to bring that work local. The drawback is memory bandwidth: at 273 GB/s, it’s the slowest of the three when a model has to move a lot of data, and reviewers consistently find it underperforms Apple’s Mac Studio on raw token generation despite the mature software.

This fits teams already deep in the CUDA and TensorRT stack who want to prototype locally before deploying to NVIDIA cloud infrastructure without touching their code. For models, think quantized versions of Llama, Qwen and DeepSeek in the 70B to 200B range, where TensorRT-LLM optimization matters more than raw bandwidth.

AMD: one chip, two very different products

This is where things get confusing on purpose. AMD’s Ryzen AI Max+ 395 chip, marketed as “the world’s smallest AI development PC” capable of running 200-billion-parameter models locally, ships in two completely different products. Third-party mini PCs built around the same chip, like the GMKtec EVO-X2 and the Framework Desktop, sell for a fraction of AMD’s own official Ryzen AI Halo Developer Platform. Same processor, same 128GB of memory and more than double the price on the official kit.

LTT Labs, one of the more rigorous independent testing outfits, explains the gap plainly: the premium buys AMD’s curated software configurations, “AI Playbooks,” and first party support, not extra hardware. The advantage of the third party route is price. You get the same chip and memory pool for far less if you’re comfortable managing the software yourself. The advantage of AMD’s own kit is exactly that support layer, plus the promise of continued driver and playbook maintenance. The drawback, either way, is raw speed: LTT Labs found “the Apple Silicon Mac Studios outperform the AMD Ryzen AI Max+ 395 machines,” generating tokens two to three times slower in their testing.

There’s a software catch too, specific to the operating system. ROCm, AMD’s answer to CUDA, is still meaningfully more mature on Linux than on Windows. Windows support runs through WSL and a newer compatibility layer that only recently expanded to more Ryzen hardware, while Linux has had direct, better supported access for longer. This fits budget conscious developers who want a large memory pool for training and fine tuning experiments on Linux, and don’t need NVIDIA specific tooling. On the models side, AMD has actually leaned into this: Ryzen AI Max+ 395 was the first consumer processor to run OpenAI’s gpt-oss 120B, which needs around 61GB of memory at MXFP4 precision and runs at up to 30 tokens per second on the 128GB configuration.

Apple: the quiet winner on bandwidth

Apple’s Mac Studio comes with the M5 Max chip and up to 128GB of unified memory, or the M5 Ultra with up to 512GB. The real differentiator is memory bandwidth, and it’s worth being precise here since the two chips aren’t equal: The M5 Max reaches up to 614 GB/s, while the M5 Ultra doubles that to 1.2 TB/s. Even the lower tier beats NVIDIA and AMD’s local options at moving model data, which is why reviewers keep clocking the Mac Studio ahead of both rivals on local inference despite Apple never marketing these chips as AI accelerators first.

The advantage is that bandwidth, plus a memory ceiling (512GB on the Ultra) that lets you load models neither of the other two machines can hold without heavy quantization. The drawback is the ecosystem: Your code needs to run through Apple’s stack rather than CUDA or ROCm, which is a real cost if your team’s tooling already assumes one of those. That said, this doesn’t mean writing everything in MLX. Most people running models on a Mac just load GGUF files directly through llama.cpp’s Metal backend, Ollama or LM Studio, all of which work out of the box. MLX only matters if you’re optimizing specifically for Apple Silicon.

This fits developers who care more about inference speed and model size than ecosystem compatibility. For models, 512GB is enough to run a 70B model with zero quantization, which is a real advantage over the other two machines. It is not, however, enough for the largest open models.

The bottom line

None of these is a universal answer. If your code already runs on CUDA, NVIDIA removes the most risk. If you want the most memory for your money and you’re comfortable on Linux, AMD’s ecosystem, official kit or third party board, gets you there cheaper. If you want the fastest actual inference and don’t mind stepping outside CUDA and ROCm entirely, Apple quietly wins the benchmarks. Pick the ecosystem your workflow already depends on first. The hardware comparison only matters after that.

A local AI box isn’t there to replace the cloud for large-scale training. It exists for three specific things: keeping sensitive data off someone else’s servers, avoiding metered API costs when you’re iterating hundreds of times a day and getting a fixed, predictable machine you fully control instead of a shared, rate-limited one. Cloud GPUs still win for large training runs and burst capacity. A local box wins for prototyping, fine-tuning smaller models and running inference on data you’re not allowed to send anywhere else. That’s the tradeoff, and it decides which of these three ecosystems actually fits your work.

NVIDIA, AMD and Apple are all shipping new local AI hardware on a near-constant cycle right now. Let’s look at which machine actually makes sense for you.

NVIDIA: the safe, expensive default

The NVIDIA DGX Spark pairs 128GB of unified memory with a Grace Blackwell GB10 chip built around CUDA, which is still the deepest, most battle tested software ecosystem in AI. Existing PyTorch code, TensorRT and cuDNN pipelines run on it without modification, which is not something you can say about the other two options yet.

My company teaches AI for business. At conferences, we show people how to save prompts they can reuse. Afterwards, someone almost always asks, why do we give these techniques away?

Two reasons: First, these techniques can be game changers in how you work. Second, everyone should know how to use AI this way — and we’re only at the beginning. After all, you have to start at the beginning, not the end.

Think about onboarding a new account manager. You would give them background on the business, examples of good work and clear expectations. You wouldn’t start that conversation from scratch every morning. Yet that’s how many people use AI.

A saved prompt captures that handoff. You explain the task, provide the context and define what a useful answer looks like. Then you reuse those instructions instead of starting over.

Let AI interview you

Start with something you do every week. A recurring task that someone quietly dreads is a strong candidate. Open a regular AI chat and paste this prompt:

I want a reusable AI assistant for [your task]. Interview me one question at a time about my role, how I do this work, what good output looks like, what I would reject and which reference files would help. Keep asking until I say we’re done. Then write clear, copy-ready prompt instructions.

AI will start asking questions, one at a time. Answer naturally, providing as much detail as you can. Use voice dictation if that helps. Half-formed thoughts and messy sentences are fine. Your job is to explain what you know; AI can help organize it.

The phrase “one question at a time” matters. It keeps the conversation manageable and lets each answer shape the next question. If the questions start to feel repetitive or your answers cover the same ground, you’re probably ready to say: “We’re done. Please prepare the prompt.”

Save the instructions where you work

Test the prompt instructions in a new chat first. Refine them until the output meets your expectations, then save them using your AI tool‘s supported features. For ChatGPT, this would be a Skill. For Claude, this is a Project. 

In ChatGPT: Select @skill-creator and ask it to turn the instructions into a reusable Skill. Give it a name and explain when it should run.

In Claude: Select Projects from the menu on the left, then choose New Project. Name your project, add a brief description and select Create Project. Next, click the plus sign next to Instructions and paste your prompt instructions there.

Save or email yourself a copy of the instructions, too. Product names and menus change. Your description of how the work should be done is worth keeping.

Try these two prompts

Replace the text in each prompt below. Paste it into a fresh chat to try it, then save it as a Project, Skill or another reusable assistant format supported by your tool.

Your professional coach

Act as my practical business coach. I am [role] at [company and industry]. Start by asking about my responsibilities, goals and biggest challenge, one question at a time. Keep asking until I say we’re done. Then summarize what you heard and ask me to correct it. When I bring you a problem, clarify the decision and constraints before advising. Challenge weak assumptions respectfully, explain trade-offs and suggest a next step with a deadline and a way to measure progress. Ask about progress when I return, using the context available. Be direct and specific. Distinguish facts from assumptions. Do not invent knowledge of my business or promise results.

Give the coach your role description and current goals. A useful first request is: “Help me decide which of these priorities deserves my attention this week.”

Your contract reviewer

Review the supplied contract from [my company’s] perspective. Ask which party we are if that is unclear. First, give a short business summary covering scope, fees, payment, term, renewal and termination. Then rank issues High, Medium or Low by business impact, with High first. Review all clauses and supplied exhibits. Check liability, indemnity, IP ownership, confidentiality, data use, assignment and missing protections. For each issue, cite the section and a short quote, explain the risk and suggest specific fallback language. Flag terms the counterparty could reasonably object to. Compare the contract with any supplied standard agreement or playbook. Separate contract facts, market judgments and proposed changes; qualify unsupported market claims. Never invent terms or legal rules. Flag missing documents, uncertainty and jurisdiction-sensitive points for counsel. Always state that the final contract should be reviewed by our attorney.

Try this prompt with your standard agreement and negotiation priorities. It can help you identify issues to raise with your lawyer and focus the review. It does not replace legal advice.

Test it on real work

Did you get perfect results? I doubt it.

Copy your saved instructions into a new chat and explain what went wrong. Ask AI to revise the instructions, then test them again. After a week, look at what you keep fixing by hand. Those repeated corrections tell you what needs to change in your saved prompt.

Use only tools and files your company permits. Check data-handling and sharing settings before uploading confidential material. Give AI examples of good work, and review every output before relying on it or sending it to someone else.

Pick one workflow and use it for two weeks. Once it becomes part of your routine, choose the next one.

My company teaches AI for business. At conferences, we show people how to save prompts they can reuse. Afterwards, someone almost always asks, why do we give these techniques away?

Two reasons: First, these techniques can be game changers in how you work. Second, everyone should know how to use AI this way — and we’re only at the beginning. After all, you have to start at the beginning, not the end.

Think about onboarding a new account manager. You would give them background on the business, examples of good work and clear expectations. You wouldn’t start that conversation from scratch every morning. Yet that’s how many people use AI.

About 33 years ago, Kelly Schultz told his friends and family that he was going to open a pumpkin farm. 

“Everybody laughed at me,” he said in a new interview with CNBC Make It. 

The endeavor was outside Schultz’s area of expertise. He was a second-generation antiques seller and didn’t know anything about farming. He bought the small farm in Clarence, New York, anyway. 

No one’s laughing at him now. 

Schultz, 74, now runs The Great Pumpkin Farm, located about 15 miles northeast of Buffalo. More than 100,000 people visit the seasonal attraction each fall.

The farm generated $2 million in revenue in 2025, according to financial documents reviewed by CNBC Make It. It operates for just seven weeks a year.

This year’s season runs from September 19 through October 31. Pumpkins are only part of the draw. Visitors can ride a pumpkin-themed train, enter pie-eating contests or explore a five-acre corn maze. Weekend admission is $20, though some activities, including the corn maze, cost extra.

How he got started

Schultz had the idea to open a farm of his own after visiting a crowded pumpkin patch in upstate New York in the early 1990s. He observed a simple operation with “muddy parking lots” and “not a whole lot of restrooms,” but there were still “a zillion people” browsing and buying pumpkins anyway. 

“I looked around, and I said, ‘I could do this, better than this,’” Schultz said. 

When the former cattle farm next to his antique store hit the market in 1993, Schultz bought the 85-acre property for about $650,000, intent on transforming it into a pumpkin farm. 

Schultz noted that more people visit the farm every year. 

“There are people that come here almost every weekend,” he said. “They love it so much, they just keep coming back.”

Americans are spending more on Halloween than ever before

For Schultz and other seasonal entrepreneurs, Halloween brings a short but lucrative window of opportunity. 

According to the National Retail Federation’s annual Halloween survey, released on September 22, Americans will spend about $13.5 billion celebrating the holiday in 2026, up from $13.1 billion last year. The survey asked 7,889 consumers about their Halloween plans.

Consumers expect to spend an average of $115.14 per person, with purchases ranging from costumes to decorations.

Halloween has become an important holiday for more people. Nearly three-quarters of consumers surveyed, or 74%, plan to celebrate this year, compared with 68% in 2019. 

Among those celebrating, 42% plan to carve pumpkins, and an identical percentage plan to decorate their homes and yards. Handing out candy remains the most popular tradition, at 61%, followed by dressing in costume, at 50%.

The sales opportunity also starts well before October 31. Nearly half of Halloween shoppers, 49%, said they planned to begin buying in September or earlier, compared with 34% a decade ago. Businesses have more of a chance to sell seasonal Halloween products weeks before the holiday itself. 

About 33 years ago, Kelly Schultz told his friends and family that he was going to open a pumpkin farm. 

“Everybody laughed at me,” he said in a new interview with CNBC Make It. 

The endeavor was outside Schultz’s area of expertise. He was a second-generation antiques seller and didn’t know anything about farming. He bought the small farm in Clarence, New York, anyway. 

It’s a strange moment to write about what makes companies last, with global uncertainty at historically high levels and leaders being asked to make huge bets on AI, talent and where the world is going. At the same time, employees are nine times more likely to be engaged when they believe their company has a secure future. 

Consider those two findings for a second: The defining tension of building a company today is that the ground is shakier than ever, and the people you lead need more steadiness than ever. That tension is exactly what the last decade of building Ninety has been about. 

I spent years as an investor and board member watching capable teams struggle with the same problems: lack of shared vision, cultures without purpose, priorities that didn’t translate into coordinated work and leaders who didn’t have a connected view of how their company was running. I founded Ninety to build the business operating system I kept wishing those companies had. In doing so, I started living these problems from the inside. 

On our 10th anniversary, here are 10 lessons I learned the hard way.

1. Vision and culture come first

That’s what attracts and retains great people. I keep coming back to the old Drucker line that culture eats strategy for breakfast. For me, that is how enduring organizations work. Ninety was never trying to be the cheapest tool or the one with the most features. It was built around service, stewardship, high trust and doing the right thing. That culture is what let us keep good people through hard years.

2. Lead with customer experience

Every company has to choose a core value proposition, whether it’s lowest cost, constant innovation, exclusivity or service. We picked service, and we’ve been tested on it. When COVID hit and revenue dried up for entire client sectors overnight, the short-term move would’ve been to nickel-and-dime people to protect our own numbers. We chose the opposite. We gave customers grace when they were struggling, because in the long run, genuine care builds trust that no marketing budget can buy.

3. Build resilience, not perfect plans

No plan survives the first real hit; what matters is whether the company can absorb it and keep functioning. That is resilience, and it comes from where you spend your energy. Energy is finite. Spend it fighting things you can’t control, like market swings, regulatory shifts or a competitor’s moves, and you’ll have nothing left for what you can control, like aligning your team, executing and managing cash.

4. Don’t burn bridges, even when you’re forced into plan B

A few years ago, EOS Worldwide dissolved its licensing program and launched its own operating system software. Ninety’s software had been built around that ecosystem, so this was a gut punch. We could have thrown a fit, taken our ball and gone home. Instead, we went back to the questions that keep our decisions aligned with who we are. We chose to stay agreeable, keep improving our own platform and remain true to our principles. When EOS later reached back out to reestablish the partnership, that grace paid off and we reengaged on far better terms.

5. Stay ahead of the curve, then be patient

You have to anticipate where the world is going and invest before customers even know they need it, whether it’s adopting technology in traditional spaces or integrating AI into core workflows. It’s important to temper your ambition with honesty, as real organizational evolution takes two to three years of grinding effort.

6. Culture has to fit the environment you’re operating in

The market has a major impact on what kind of culture you need. A fast-moving, highly ambiguous space, like what we see around AI right now, requires people who can live in the gray area, move quickly, work hard and adapt. A culture that was perfectly right for one environment can start to break when the environment changes.

7. Leadership demands evolve with the organization

Every stage of development asks something different of the people leading it. Some leaders keep expanding their capacity to handle complexity, while others hit a ceiling they can’t even see. For example, the right team at 10 people might not be the team that carries you through to 100 or 300 employees. Someone can be a wonderful human and strong cultural fit and still not have what the next, more complex seat requires. Building something enduring means telling the truth about what the company needs.

8. Own the one-way-door decisions yourself

Founders make plenty of bad calls. The point is to stay humble enough to see what happened, own it and get better. But some decisions can’t be undone, like letting people go. Those one-way-door moments can reveal a lot about the true nature of the organization and its leaders. You can delegate a lot of tasks and responsibilities as you scale, but you can’t fully outsource decisions that carry deep cultural consequences. Those you have to carry yourself and make sure they align with your values.

9. Build mechanisms for the truth to surface

Great companies don’t have bystanders. The thing that goes unsaid is usually the thing that eventually hurts you. Enduring organizations create shared language and enough trust that people feel comfortable raising issues before they become permanent problems. At Ninety, we lean on phrases like “succeed or escalate,” “what’s working, what’s not,” and “what are the growth opportunities?” They may sound simple, but each one gives someone permission to surface a concern early.

10. Remember that everyone’s job is to help the organization win

It takes more than everyone simply doing their specific function well or optimizing their own department. Everyone has to understand what winning looks like and be able to step outside their lane when the company needs it. That is where a vision, a handful of clear quarterly rocks and an integrated business operating system earn their keep: They turn scattered priorities into coordinated work that everyone can see.

Closing thoughts

Ten years in, with more than 18,500 companies now running on Ninety, I’m more convinced than ever that building a great organization is a beautiful mess. It’s hard, humbling and full of lessons you would never have chosen. Love the game, be great at the work, stay humble, show grace and keep getting better. If the creek don’t rise and the sun still sets, you’ll find a way forward and be stronger for it in the end.

It’s a strange moment to write about what makes companies last, with global uncertainty at historically high levels and leaders being asked to make huge bets on AI, talent and where the world is going. At the same time, employees are nine times more likely to be engaged when they believe their company has a secure future. 

Consider those two findings for a second: The defining tension of building a company today is that the ground is shakier than ever, and the people you lead need more steadiness than ever. That tension is exactly what the last decade of building Ninety has been about. 

I spent years as an investor and board member watching capable teams struggle with the same problems: lack of shared vision, cultures without purpose, priorities that didn’t translate into coordinated work and leaders who didn’t have a connected view of how their company was running. I founded Ninety to build the business operating system I kept wishing those companies had. In doing so, I started living these problems from the inside. 

AGI, artificial general intelligence, means a system that can match or beat human performance across essentially any cognitive task, not just the narrow ones it was trained for. That distinction matters because it’s exactly what’s being argued about right now.

Jensen Huang has a habit of declaring victory at the exact moment his company needs the market to believe it. So when OpenAI shipped GPT-6 Astra on Sept. 3, Huang posted on X that “AGI has arrived,” adding that “400K GPUs” are “coming online next.”

I’m an AI engineer, and I’m currently building ChatMakr, an AI-powered platform that creates multi-agent chatbots in seconds. I spent the last year watching every model release get relabeled as the one that finally changes everything. Here is the truth.

What Astra actually does

Astra isn’t a chatbot upgrade. OpenAI positions it as an operator — a system that browses, codes, runs multi-step workflows, models in Blender and handles cybersecurity tasks with minimal supervision. The benchmark numbers are genuinely striking: 97.6% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and a perfect 100% on ExploitBench, up from 78.5% for its predecessor, GPT-5.6 Sol. Those are not incremental gains.

But OpenAI itself, notably, stopped short of calling Astra AGI, choosing instead to emphasize “speed and versatility.” That restraint from the company that built the model is the tell.

Why people aren’t convinced

The pushback isn’t really about whether Astra is impressive. It clearly is. It’s about the word AGI itself. Wharton professor Ethan Mollick has argued for months that what we’re seeing is “jagged AGI:” systems that are superhuman in some narrow domains and unreliable in others, which is a very different thing from general intelligence.

There’s still no agreed upon test for AGI, so declaring it “arrived” is closer to a subjective judgment call than a measurement. Augmented Mind and Manolo Remiddi put it plainly: “AGI is the point where you can trust the output without checking. We are not there.”

The Fable 5.1 comparison and why scores lie

This is where it gets interesting for anyone actually building with these models. On paper, Astra and Anthropic’s Claude Fable 5.1 look neck and neck on several coding and reasoning benchmarks. In practice, engineers report meaningfully different output quality. Fable tends to produce more production-ready code with fewer output tokens, while Astra’s strength shows up more in computer use and multi-step automation. Snorkel AI’s testing revealed that Fable fails through a fundamentally different pattern than how it succeeds, losing valid work mid-trajectory rather than running out of budget, which matters far more to a shipping team than a leaderboard percentage.

The reason similar scores can hide such different outputs comes down to something most people outside AI engineering have never heard of: the harness.

What a “harness” actually is

A harness is the infrastructure wrapped around a model: the prompt assembly, tool access, memory management and error recovery that determine how the raw model is actually put to work. As MindStudio describes it: “A model is the engine. The harness is the car. Most benchmarks test the car.” Swap the harness and the same model produces wildly different results. Cursor’s own research found one model scoring 46% accuracy with one harness and 80% with another, with no model change at all.

Astra’s own benchmark history makes the point better than any explainer could. On ARC-AGI-3, Astra scored 62.7% using a standard harness, essentially tied with its predecessor, and 99.9% using a “Provider Adapter” harness that let it retain and reuse prior reasoning instead of restarting cold on every step, according to Augmented Mind and Manolo Remiddi. Same weights. Same model. A 37 point swing from infrastructure alone. That’s not a small footnote. It means a huge share of what gets reported as “model improvement” is actually harness improvement, and leaderboards that don’t control for it are telling a story, not reporting a fact.

Why Cursor just became the most important acquisition of the year

This is exactly why SpaceX’s $60 billion acquisition of Cursor matters more than it looks. Cursor isn’t a model company. It’s one of the best harness companies in the world, the layer that decides how coding models actually get used. Folding it into SpaceX’s GPU fleet is a bet that owning the layer is as valuable as owning the model, maybe more. OpenAI has already cut Cursor’s access to its models over trust concerns tied to Musk’s companies, which tells you the model providers see the same thing: Whoever controls the harness controls the product experience, regardless of which model sits underneath it.

The bottom line

Nobody serious disputes that GPT-6 Astra is a major model. What’s contested is the leap from “major model” to “AGI has arrived.” If you’re evaluating these systems for your own business, ignore the AGI framing entirely. Look at the harness a vendor wraps around the model, ask how it fails, and test it on your actual workflow. The benchmark gap between Astra and Fable 5.1 already proved the leaderboard alone won’t tell you.

AGI, artificial general intelligence, means a system that can match or beat human performance across essentially any cognitive task, not just the narrow ones it was trained for. That distinction matters because it’s exactly what’s being argued about right now.

Jensen Huang has a habit of declaring victory at the exact moment his company needs the market to believe it. So when OpenAI shipped GPT-6 Astra on Sept. 3, Huang posted on X that “AGI has arrived,” adding that “400K GPUs” are “coming online next.”

I’m an AI engineer, and I’m currently building ChatMakr, an AI-powered platform that creates multi-agent chatbots in seconds. I spent the last year watching every model release get relabeled as the one that finally changes everything. Here is the truth.

This Side Hustle Spotlight Q&A features Georgia Young, a 28-year-old entrepreneur based in New York. Last year, Young was working in consulting when she started her side hustle Rallee: a line of caffeinated gummies for energy on the go. Rallee landed in 1,940 Walmart stores in eight months and is projected to bring in about $1.8 million in revenue in 2027. Responses have been edited for length and clarity. 

What was your day job or primary occupation when you started your side hustle?
Before starting Rallee, I spent four years in strategy and sustainability consulting, advising private equity and private credit firms, global financial institutions and asset managers on their sustainability and impact investing strategies across multi-billion-dollar portfolios. It was an incredibly demanding environment with tight deadlines, overlapping deliverables and long hours.

I loved the pace and the challenge, but my days were often completely dictated by client demands. My background was also very far from consumer packaged goods. I have a master’s degree in sustainable business from the University of Miami. So when I first started working on Rallee, I was essentially teaching myself an entirely new industry while balancing a demanding full-time job.

A demanding 9-5 job inspired her side hustle Rallee

When did you start your side hustle, and where did you find the inspiration for it?
I developed Rallee while I was still consulting and launched it in January 2025. The inspiration came directly from that job. Working with demanding private equity and private credit clients meant constant overlapping deadlines, back-to-back meetings and very long hours. I was always trying to find enough energy to keep up, both professionally and personally.

Like so many people, I relied on caffeine, but I was increasingly frustrated with the options available. I didn’t have time to stand in line for a barista every time I needed a boost. Spending $5 or $6 on coffee or matcha several times a day added up quickly. And I wanted something better for me than traditional energy drinks. I wanted caffeine that was more affordable, convenient and thoughtfully formulated.

Researching manufacturers and testing samples

What were some of the first steps you took to get your side hustle off the ground? How much money/investment did it take to launch?
The first step was figuring out if the product I had imagined could even exist. I researched gummy manufacturing, caffeine delivery systems and ingredient interactions. Then I worked with nutritional scientists to develop a formula that was effective, jitter-free and enjoyable to eat. We paired caffeine with L-theanine to support calm focus without the jitters; used a coated, slow-release caffeine for steadier energy instead of a spike; and added cordyceps mushrooms, all in a vegan, 10-calorie gummy made with real fruit.

From there, I reached out to manufacturers, tested samples and learned about regulatory requirements, packaging and branding. I had absolutely no background in food manufacturing. So the learning curve was huge. Because I was doing all of this alongside my consulting job, evenings, weekends and pretty much every spare moment went into getting Rallee off the ground. I kept consulting until April 2025, about three months after launch, before going full-time. Very quickly, I learned that a physical product is expensive before you’ve sold a single unit because formulation, MOQs, packaging, testing and regulatory work all add up fast. I invested all of my savings, around $45,000, to launch Rallee, and the business is still 100% founder-owned and self-funded today.

Tapping into advice from other founders

Are there any free or paid resources that have been especially helpful for you in starting and running this business?
Other founders have been my most valuable resource by far. I’ve learned an enormous amount from people who are a few steps ahead of me, particularly on manufacturing, retail negotiations, fundraising and distribution. You can spend hours researching something online, or you can have a 20-minute conversation with someone who has already been through it. Communities like Startup CPG have also been incredibly valuable. One of our biggest opportunities came through a Walmart submission connected to the community, which ultimately led to Rallee securing national distribution with Walmart. 

If you could go back in your business journey and change one process or approach, what would it be, and how do you wish you’d done it differently?
I would have been much quicker to accept that not everything needs to be perfect before you launch. Coming from consulting, I was used to deliverables going through multiple rounds of review before they ever reached a client. I wanted every detail to be exactly how I had envisioned it. I’ve learned that building a consumer brand is an ongoing process of testing, learning and improving. You can spend months debating something internally, only to discover that customers care about something entirely different. 

That doesn’t mean compromising on quality or safety, but perfectionism is incredibly expensive when you’re funding a business yourself. Getting to 90% and learning from the market is often far more valuable than spending another six months chasing the final 10%.

Building a physical product requires significant cash

When it comes to this specific business, what is something you’ve found particularly challenging and/or surprising that people who get into this type of work should be prepared for, but likely aren’t?
The thing that surprised me most is how much cash a physical product ties up before you ever see a sale. Every production run means paying upfront for gummies, packaging and freight, often months before that inventory turns into revenue, and retail stretches that gap even further. On top of that, people underestimate how many different hats you have to wear, especially early on when you don’t have the budget to hire an expert for every problem. One minute you’re the creative director, the next you’re the finance team, and by lunchtime, you’re troubleshooting a manufacturing issue. It’s challenging, but it’s also one of my favorite parts of being an entrepreneur.

Can you recall a specific instance when something went very wrong? How did you fix it?
One example was during production for our Walmart launch. We were working toward a major retail deadline when a piece of sealing equipment we needed for our packaging didn’t arrive in time, and pushing production back simply wasn’t an option because we had retailer deadlines to meet. So, we completely rethought the process. We brought in additional labor, switched to handheld sealing equipment, and worked closely with our manufacturing team to get everything packed and ready. It was significantly more labor-intensive and expensive than we had planned, but we found a way to make it work.

7.6X increase in DTC sales, on track for $1.7M next year

How long did it take you to see consistent monthly revenue? How much did the side hustle earn?
Honestly, it took longer than I expected, and I think that’s one of the most useful things I can share with anyone starting out. We launched in January 2025 and had sales from the very first month, but they were inconsistent through most of that first year while we worked out the right format, pricing and messaging. The turning point was our February 2026 relaunch. Once we focused on subscriptions, bundles and single-serve sampling pouches, our direct-to-consumer sales grew 7.6 times in four months. Almost everything we earn still goes straight back into the business because product development, inventory, packaging, marketing and logistics all require continuous investment. The distinction between revenue and profit is really important for anyone building a physical product business.

What does growth and revenue look like now?
Since our February relaunch, Rallee has gone from a direct-to-consumer brand to 1,940 Walmart stores in eight months, with our new six-gummy mini variety pack rolling out nationally and our full-size tubes launching in select stores. We’re conservatively projecting about $1.8 million in revenue in 2027. Our focus now is on expanding our retail presence, growing brand awareness, developing new formats and building a business that can scale sustainably. For me, the most exciting part is that we’re still at the very beginning of what Rallee could become.

What to do before spending thousands on product development

What do you enjoy most about running this business?
I love seeing something that started as an idea in my head become a real product that people genuinely love. There’s something incredibly rewarding about watching someone pick up Rallee for the first time, try it and understand exactly why I created it. One customer told us it was the first caffeine product that didn’t make her anxious, and that is exactly the problem I set out to solve. I also love the creative freedom of building a brand from scratch, developing new products, making decisions and seeing those ideas come to life, which is something I never take for granted. And of course, seeing Rallee on the shelves at Walmart was a pretty surreal moment. It’s one thing to believe in your own idea, but it’s another to see it sitting alongside brands you’ve known your entire life.

What is your best piece of specific, actionable business advice?
Before you spend thousands of dollars developing a product, speak to at least 20 people who you genuinely believe would buy it, and don’t steer them toward the answers you want to hear. Ask them what they’re currently using, what frustrates them about it, how often they buy it and what would actually make them switch. Their answers should shape your product, pricing and messaging before you’ve committed to your first production run.

This Side Hustle Spotlight Q&A features Georgia Young, a 28-year-old entrepreneur based in New York. Last year, Young was working in consulting when she started her side hustle Rallee: a line of caffeinated gummies for energy on the go. Rallee landed in 1,940 Walmart stores in eight months and is projected to bring in about $1.8 million in revenue in 2027. Responses have been edited for length and clarity. 

What was your day job or primary occupation when you started your side hustle?
Before starting Rallee, I spent four years in strategy and sustainability consulting, advising private equity and private credit firms, global financial institutions and asset managers on their sustainability and impact investing strategies across multi-billion-dollar portfolios. It was an incredibly demanding environment with tight deadlines, overlapping deliverables and long hours.

I loved the pace and the challenge, but my days were often completely dictated by client demands. My background was also very far from consumer packaged goods. I have a master’s degree in sustainable business from the University of Miami. So when I first started working on Rallee, I was essentially teaching myself an entirely new industry while balancing a demanding full-time job.