The Miami AI Dossier

A · Bottom line

The excitement probably came from the hallway, not the stage

The best-fitting event is The Real Deal Miami Real Estate Forum, Nov 4–5 at Mana Wynwood. It drew about 6,000 people and ran more than three dozen panels. Only two were about AI, and neither touched AI video for agents. A man selling AI avatar video to agents had a room full of top agents to pitch, which is a better reason to go than the agenda.

• The eventThe Real Deal Miami forum is the best fit, now that we know he serves real estate agents. This is still inference: nothing public shows he attended.
• The ideaAI that does routine work for you: cloned-presenter videos, follow-ups, content at scale. The tools got much cheaper this fall.
• His offerReal but crowded. Avatar tools cost $24–$149 a month, HeyGen and big brokerages now offer their own versions, and agent video retainers cluster at $199–$499 a month.
• Leonard's ruleSound, for a different reason than surprise breakage. Vendor changes now come on a schedule, but each one is still real rebuild work, and a flat maintenance fee leaves the builder either absorbing it or upselling to cover it. The crowd selling AI setups is real too, though it's crowded among sellers, not buyers.

C · Which event

The week fell in a gap between Miami's AI events

The dedicated Miami AI Summit ran on Oct 23. The next AI-heavy wave, including the TBM Conference and Adobe MAX, starts Nov 9–10. Business events in between ran AI as a side topic. Once we learned his clients are real estate agents, one candidate pulled clearly ahead.

D · What the forum covered

Capital, condos and politics, with AI in two sponsor slots

The main stage covered private equity, the Miami market reset, global buyers, aging condos, Miami Worldcenter, and Ken Griffin's $1.1 billion purchase of the Mana Wynwood site. Mayor Daniella Levine Cava spoke, and Rick Ross made an appearance about his Miami Gardens project. The press release announcing the forum never mentioned AI.

Both AI sessions ran on Day 1 on the secondary Innovation Stage, where every slot was sponsor-presented:

AI has also been moving down the bill each year:

No outside press, Reddit threads, LinkedIn posts or recordings of the 2026 edition turned up. The forum ended only two to three days before this research, so posts may still appear. The closest thing to an inside account of a recent edition is a 2025 sponsor blog, which said agents were moving toward "automation… and AI tools that allow them to scale without burning out," and that authentic video was winning trust.

E · His business

A media company whose AI work happens off its websites

Soare Productions' public sites describe a traditional real estate media and social-media company: photography, cinematic 4K and drone video, Matterport tours, virtual staging, agent intro videos, reels and social-media management. Its packages are called Foundation, Momentum and Ascend, with no published prices. A second search on Nov 7 covered every site the company links to: the main site, its Hawaii and Los Angeles pages, soareintl.com, soarebc.com and soarealberta.com. It also searched the founder's name and the brand for any separate AI business. None of it mentions AI avatars, digital twins, AI implementation, training or retainers, and no separate AI brand turned up. That fits what he has shared privately as a long-time client: the AI side is sold through direct relationships and conversations rather than advertised online. His company Instagram (@soareproductions, about 11,300 followers) is the one public place the shift shows. Its bio now reads "We Help Realtors Grow Through AI and Social Media," claims work with over 1,500 agents, and links to a booking page for a free intro call. Instagram shows little else without signing in, so the details of the AI offer still come from what he has told your partner. His websites undersell what he now does.

His personal Instagram (@cipyy, about 8,600 followers) shows where he is taking the business. The bio now reads "Real Estate Personal Brand Architect," helping agents "attract sellers through content, story & IG systems." His pinned posts from late 2025 describe the pivot in his own words: "Before I ever coached an agent, I spent 10+ years behind the camera building the brands of over 1,000 realtors… Now I teach those systems." The pitch is about positioning, not volume: "no dancing, no posting 3× a day, no 'gaming the algorithm,'" and "It's never about the volume. It's always about the positioning." His way in is a free Instagram audit for any agent who messages him "AUDIT." His biggest recent posts aren't sales posts at all. A June reel asking people to tag local Hawaii businesses drew about 6,200 likes and 600 comments. None of the captions checked mention AI by name.

Read together, the two accounts suggest the AI work is the production engine behind a coaching business. The clone and the workflows let an agent stay visible without filming constantly, and the brand strategy is what he sells. That is a more defensible position than reselling avatar tools, and it is closer to the "outcome plus insurance" offer described later in this report than his websites suggest.

The existing business gives him real advantages for the new one. He already has agent clients, already films the footage avatar training needs, and already runs social-media management, which is a retainer by another name.

The competition is closing in from two directions. In August, HeyGen launched its own real estate product: record once, then generate market updates and listing spotlights. Keller Williams added Rejig.AI, which makes agent-branded posts and videos automatically when a listing goes live, and eXp is expanding its own AI platform. On Oct 13, SERHANT. started a daily avatar channel with HeyGen, with no results yet. Evidence that avatar video wins agents business is still thin. The best success story found is one agent who closed two deals over five months, and he sells a cloning tool.

Verdict: teaching agents to clone themselves is the weakest part of the offer, because HeyGen gives away the playbook. Done-for-you production with distribution and compliance holds up better, at squeezed prices. AI implementation for general businesses is a bigger market with better margins.

F · His numbers

An outside estimate: $0.8–2 million a year in gross revenue

Soare is private and publishes no prices, so this is built from his own public claims and typical market rates. About $0.5 million is a realistic floor, and $3 million would need every service line near its high end at once. Treat it as a guess, not something to repeat to him or others as fact.

The payroll gap is the telling detail: "lots of employees" and 6–10 on payroll can both be true if most of the team are contractors paid per job. No public source gives a filed revenue figure. Dun & Bradstreet, Crunchbase and PPP loan records turned up nothing.

Open the Revenue estimate tab →

G · Does it all break?

Changes come on a schedule, and every one is still real work

Leonard's concern isn't that things break without warning. It's dependency: a business built on a few AI vendors inherits their change schedule, and a published schedule doesn't make the rebuild work go away. The 2026 record bears that out, with seven changes in seven months that forced someone to migrate, re-test or rebuild:

1. 2026-05-15xAI retires the original Grok 4 API modelAbout nine days' notice; requests quietly routed to a different model at different rates.
2. 2026-06-15Claude Sonnet 4 and Opus 4 retiredAnnounced Apr 14, about two months ahead.
3. 2026-06Zapier moves AI steps to tiered pricingUp to 5x the cost for some AI steps.
4. 2026-08-05Claude Opus 4.1 retiredAnnounced Jun 5.
5. 2026-09-24OpenAI shuts down the Sora APIThe app closed in April. Anything built on it had to move.
6. 2026-10-21Anthropic API drops the temperature settingAny no-code AI step that hard-codes temperature now returns an error.
7. 2026-11-30Claude Sonnet 4.5 retires23 days from this report. Every workflow still on it needs a swap and a re-test.

Newer agent tools change the kind of upkeep more than the amount. Grok Bot is a real xAI product, in beta since Aug 11: agents that run on their own cloud computers and operate tools like a person. Claude Managed Agents and AWS's agent service do similar work, and shared standards like MCP and Agent Skills make tool connections portable across models. A goal-driven agent recovers from a renamed field far better than a rigid chain of steps. In exchange you get beta products that change, costs that spike when an agent loops, and new security holes. The open-source OpenClaw had remote-code and prompt-injection flaws, and Cisco found a third-party skill sending data out.

The work has shifted from fixing broken steps to watching, securing and tuning a semi-autonomous system. It breaks less, but it still needs an owner.

Pricing the dependency without manufacturing upsells

Leonard's objection to the common pitch holds: a $5,000 setup plus $1,000 a month, followed by surprise "quarterly upgrades," feels manufactured, because the client couldn't see the extra work coming. The honest version puts the dependency in the contract instead of hiding it in a flat fee:

• Vendor accountsThe client owns the AI and automation accounts and pays those bills directly, so they can see exactly which companies their system depends on.
• The retainerCovers monitoring, small fixes and a stated number of hours a month. Nothing more is implied.
• Vendor changesNamed in the contract as separate, billable change work at a set rate, with a calendar of known retirements, such as Sonnet 4.5 on Nov 30. The client sees each one coming.
• DesignBuilt so a model swap happens in one place and automatic checks re-test the output. Each migration becomes hours, not weeks, which keeps the change bills small.
• Or no retainerA one-time build with a full handover, and migrations quoted only when the client asks for them.

The difference from a manufactured upsell is disclosure. Extra work the client was told about on day one, caused by a vendor they can see, is a known cost of the tool. Extra work that appears as a surprise is the pattern Leonard is right to avoid.

H · Who pays, and why

Anybody can do it, and everybody is selling it

Look from the customer's side and the work changes shape. An owner who can't build the workflow also can't spot a silent quality drop, migrate before Nov 30, or close a security hole. That work is real, so it can be billed honestly, as long as it's priced openly rather than buried in a flat fee.

Goldman Sachs small-business survey, March 2026, except where noted. The gap between the top bars and the bottom two is the done-for-you market.

The crowd is real, but it's crowded on the sellers' side

Leonard is right that the seller side is packed. Agencies, course graduates and teenagers all sell the same AI setups and agent systems. The research found the same thing: the low end is full of generic chatbots and copy-paste automations, and the marketing is "extremely noisy with people selling the dream." Most people who start never reach meaningful revenue. Bigger players are moving in too. HeyGen and brokerages like Keller Williams and eXp now bundle the tools themselves, and IT service firms, a market of about $650 billion, are adding AI services. That crowding is why agent video retainers have sunk to $199–$499 a month.

The buyer side looks different. Only 14% of small businesses have fully built AI into their operations, and fewer than one in five had paid for an AI tool at all. The market isn't full. The pitch is. When every owner gets the same cold offer, they stop trusting the offer, and the scarce thing becomes trust: an existing relationship, a track record in one niche, and proof from a client they know.

That's why your friend is better placed than the crowd. He didn't start by selling AI. He spent a decade earning the trust of 1,000+ agents with video work, and now offers AI to people who already know him. For anyone without that list, the way through is the same on a smaller scale: one niche, a few clients who already trust the builder, and published results, instead of another cold pitch into the noise.

The market has priced this in. Small-business AI work typically sells as a $2,000–$15,000 build fee plus a $500–$3,500 monthly retainer. These are agency rate cards, not audited figures. Agent video retainers at $199–$499 sit at the bottom of that range, which shows how commoditized that one deliverable has become.

His skepticism still has grounds. Operators report clients leaving once the novelty fades, and clients asking why they pay to maintain a tool they can see running on its own. Gartner expects more than 40% of agent projects to be cancelled by the end of 2027.

Has it cost Leonard? Yes, at the margin. Someone who started reselling avatar setups or chatbots in 2024 has probably earned money Leonard didn't, but in the segment now being squeezed. The bigger cost is time in the market: the client list, case studies and referrals your friend is building now.

Where precision gets paid

Four views of the same market, to show where Leonard's standards are an advantage and where the crowd has already set the price.

1 · Who has adopted AI so far

Share of firms using AI, Census Bureau survey, wave ending May 3, 2026.

Construction and food service rank below retail in Ramp's spending data (Sept 2026), but no Census figure is published for them. By size, 37% of firms with 250+ employees use AI, against under 20% of firms with four or fewer. The gap sits in small, hands-on businesses, which is where a careful builder has the least competition from in-house teams.

2 · What's stopping owners

Share of small-business owners, from three separate 2026 surveys. Bars from different surveys aren't directly comparable.

Teal bars are barriers; gold bars measure trust. Owners don't mainly lack tools. They lack someone they trust to choose, set up and check them. Accuracy and brand voice are exactly where careless work shows, and where his attention to detail is the product.

3 · Where the price floor sits

Typical monthly prices from the research. Gold bars are self-serve tools; teal bars are done-for-you services.

Where a self-serve tool does the job, a service gets priced against the tool, which is what squeezed agent video retainers to $199–$499. The wide teal bar at the bottom is custom work that no subscription replaces. That's the band where quality can set the price.

4 · Crowding against stakes

Positions are judgment calls drawn from the research, not measured data.

The sweet spot is where errors cost the client real money or legal trouble, buyers can tell good work from bad, and the crowd's copy-paste approach fails.

Where Leonard's traits pay, and where they cost him

Leonard already builds estimating, invoicing and client-communication systems for his own contracting business. That's the open, high-stakes corner of the chart, in a sector that ranks near the bottom for AI adoption. The gaps to close are the ones the crowd is good at: shipping a first version before it's perfect, showing the work publicly, and collecting proof from a few clients. Those are where the crowd wins today, not on skill.

I · A better offer

Sell an outcome plus insurance, not a tool and a tutorial

The high-quality version is mostly engineering discipline that resellers skip, which is where Leonard's high standards become an asset.

The best clients already have a working sales process for the system to plug into. Even Alex Hormozi warns that an AI sales agent dropped into a business without one performs badly.

If Leonard tests this, the advice is to skip avatar training. Pick one niche, sell build-plus-retainer at the agency price levels, and design for replaceability from day one. Your friend is a natural first conversation: he has agent relationships and footage, and Leonard has the engineering his offer lacks. Run one client for 90 days with a lead or hours-saved number agreed up front. That result would show whether Leonard's skepticism has been protecting him or holding him back.

J · Questions to ask

Curious questions that separate substance from pitch

K · Sources and limits

What this rests on

Two rounds of web research on Nov 7, 2026. Limits: no recordings or write-ups of the two AI sessions exist, so what was said in them is unknown. Social posts about the forum were not yet indexed. His social media and direct sales materials were not reviewed. Some figures (Gartner, agency rate cards, tool prices) come from secondary or vendor sources.

• TRD Miami forum schedule
• The Real Deal forum recap, Nov 6
• TRD forum press release
• TRD on trd.ai, 2023
• TREMGroup 2025 forum recap
• Refresh Miami on WIRED World Fair
• Soare Productions
• HeyGen real estate launch
• Real Estate News on KW and eXp AI tools
• HousingWire on SERHANT.'s avatar channel
• NAR 2025 technology survey
• Guide to agent video retainer pricing
• Anthropic model deprecations
• xAI Grok 4 nine-day notice
• Agent Skills donated to the Agentic AI Foundation
• OpenClaw security issues
• Goldman Sachs small-business AI survey
• Gartner on agent project cancellations
• AI agency pricing (Forge)
• AllAble on agency churn

A · The estimate

Most likely $0.8–2 million a year in gross revenue

That range comes from adding up each service line at typical market prices, using his own public claims for volume. About $0.5 million is a realistic floor, and $3 million is a stretch that would need every line running near its high end at once.

B · Public records

No public source gives a real revenue figure

Private companies in Canada and the US don't publish financials. Business databases like Dun & Bradstreet, ZoomInfo and RocketReach show modeled guesses, not filed numbers, and all three blocked automated access behind bot checks, which I didn't try to get around. What the searchable records show:

Database figures come from search result summaries, not the full profiles. If you or your partner have a ZoomInfo, Apollo or D&B login, their modeled revenue figure would be a useful second opinion, though those models are often wrong for small firms.

C · How it adds up

Four service lines, built from his own claims

D · Signals

What pushes the number up, and what holds it down

The headcount gap is the most telling detail. "Lots of employees" and a payroll of 6–10 can both be true if most of the team are contractors paid per job, mainly offshore editors. That model is efficient, but it's smaller than it sounds from the outside.

E · Revenue vs profit

Gross revenue isn't what he keeps

Editors, shooters, travel, software and AI tools could take 40–60% of revenue in a business like this. Using the middle of the range as an example:

That would be a very good business for one founder. It's probably smaller than "operating at a high level" suggests, and clearly bigger than a solo videographer.

F · What to listen for

Details that would move the estimate

If any of these come up in conversation, they'd narrow the range more than any database could.