Daxos Research: Dispatcher, Rated

1. The headcount contradiction resolves against Harmonic. The real team is three people plus one advisor, not 28, and no unrecorded capital is needed or found, so the only datapoint that made this company look unusual was false.
2. The team is materially better than a $500,000 accelerator cheque usually buys: an early Paladin Drones autonomy engineer who demoed this product publicly before incorporating, paired with an ex-Anduril counter-UAS perception PM and former F/A-18 pilot, who together pulled a Sentry SWE II out of a Series E company.
3. Everything commercial is zero. No customer verified by anyone but the company, no federal dollar across 207,731 SBIR award rows and usaspending, no FAA waiver across 1,595 records, two App Store ratings, and a marketing page frozen for eleven and a half months.
4. The wedge is transitional, not durable. "Turns any drone into an agent" means DJI in practice, per the company's own App Store listing, and that is the one half of the US fleet the FCC closed to new units on 2025-12-22, while the growing NDAA-compliant half ships bundled autonomy behind gated SDKs.
5. COMPANY 5.67, DAXOS 4.5, VERDICT MEET: there is a forced financing event inside four months and three artifacts that would settle the whole thesis in one hour, so take the call, but the default outcome of that call is PASS.

The one artifact that matters: a real shipped iOS app

So what is marketed is a lights-out 24/7 autonomous site-security system. What ships is a pre-1.0 iPad and iPhone command console that flies DJI aircraft with an operator watching a streamed feed. That gap is the whole investment question.

Harmonic's 28 is a LinkedIn associated-members artifact. The real team is three plus one advisor. No unrecorded capital exists.

The series is noise, not employment. Harmonic holds 43 datapoints from 2025-10-04 to 2026-09-30. It runs 8, then 10 through early 2026, then 12 on 2026-05-13, 6 on 2026-05-21, and 13 on 2026-06-06. A company this size does not halve its staff in eight days and more than double it back eight days later. It then runs 16 through August and jumps 19, 22, 28 across September while LinkedIn followers moved only 276 to 311. Twelve real hires do not produce a 35-follower change.

The department split gives it away. headcount_operations equals 28, the entire headcount, with zero in engineering, product, data or design. A 28-person drone autonomy company with no engineers is not a company. A truck dispatcher, a 911 dispatcher and a scrap-yard dispatcher all classify as Operations by title.

The roster is freight dispatchers. The full 30-record associated-member roster of linkedin.com/company/dispatchertech was scraped 2026-10-05 and LinkedIn's own pagination reports totalElements: 30. It contains members titled "Dispatcher" in Sri Ganganagar, Indore, Addis Ababa, Islamabad, Astana, Ludhiana, Bishkek and Georgetown Guyana, one member literally named "Iconic Dispatcher", one named "trucking industry", one title written "Диспетчер" and one typed "Daoatcher". The single associated member LinkedIn exposes to a logged-out visitor is a scrap-metal recycling dispatcher in Stanton, California, recently laid off, with 5 connections and the job title "Dispatcher". Harmonic's own keyword search returns 7,067 companies for "Dispatcher".

The company and its investor both say single digits. The LinkedIn page's self-declared size band is "2-10 employees". a16z's own portfolio page says 2 employees. Per house rules neither is the headcount of record, but both are evidence about what Harmonic's 28 is measuring.

Funding per head is therefore about $167,000, not $17,857. The "implausible" tension in the brief dissolves.

Two Harmonic data errors, recorded for future reads

Harmonic returns two "Founder" records that are the same man: person 82947044 (in/maxwelllwang, full history, current) and person 197370029 (in/maxwell-wang-7147a5392, no education, no outside experience, not current). Maxwell Wang keeps two LinkedIn profiles and Harmonic ingested both. As a result Harmonic's people array misses Nick Clouse and Kyle Mumma entirely while counting 28 phantom staff. Coverage here is incomplete in both directions. This also corrects the brief: Harmonic does not hold zero person records, it holds three, one of which is a duplicate.

The unrecorded-capital hunt came back empty, and it was run properly

No record found of any funding beyond the single $500,000 round. SEC EDGAR full-text search (efts.sec.gov, 2026-10-05, User-Agent set): "Dispatcher Systems" zero relevant hits, "Dispatcher Inc" filtered to Form D zero hits, "Nick Clouse" zero, "Nicholas Clouse" zero, "Maxwell Wang" two unrelated hits (Cabaletta Bio 2022, Cellular Biomedicine 2013) plus mutual-fund proxy tables. EDGAR browse-edgar for company=dispatcher, type=D returns "No matching companies". There is no Form D for this company. Harmonic funding_rounds is an empty array, currently_raising is null and num_funding_rounds is 1, all re-confirmed today. No press anywhere claims a round Harmonic does not record, so there is no competing web figure to mark unverified.

The $500,000 reconciles to the dollar. a16z Speedrun publishes one non-negotiable deal: $500K upfront for 10% via SAFE at acceptance, plus a further $500K into the next round within 18 months on the same terms as the next lead, which is where the "up to $1M" headline comes from (speedrun.a16z.com/faq, read 2026-10-05). Harmonic's $500,000 dated 2025-07-28, typed ACCELERATOR_INCUBATOR, sole investor a16z speedrun, is exactly the upfront half. The second half is a right, not cash, and it needs a third party to lead and price a round inside a window that closes around 2027-01-28. That is the most actionable date in this file.

Discard Harmonic's estimated $3,000,000 valuation. Harmonic's own field flags itself source: ESTIMATED, "Estimated from $500K ACCELERATOR_INCUBATOR round using typical 20% dilution". Speedrun's published standard is 10% for $500K, implying roughly $5M post-money. Harmonic valuation is null. Do not quote $3M as a valuation to anyone.

What this does to the rating, both directions

It removes the alarming reading, which was a company burning a secret raise. It also removes the flattering reading, which was a company that attracted 28 people on $500K. What is left is an ordinary three-person pre-seed with one accelerator cheque, exactly what the funding record says. On net the company becomes less interesting, not more, because the anomaly that made it worth a second look was a scraper artifact. Everything below rates a three-person company.

Maxwell Wang, co-founder and CTO Harmonic person 82947044

UIUC BS and MS Computer Science. He is roughly 25: GitHub repos APCSWORKSPACE and workspace-apcs pushed 2017 and Lynbrook High School film club 2018 to 2019 put him in the class of about 2019, so he was at Paladin as a roughly 19-year-old undergraduate. That is squarely the young technical founder Speedrun selects for.

He is genuinely technical, and the hard evidence is a third-party demo, not GitHub. He gave a public talk and live demo of Dispatcher at AI Tinkerers Palo Alto on 2024-11-21, written up 2025-01-06, described there as "SpaceX software engineer and creator of Dispatcher" (one-shot.aitinkerers.org, read 2026-10-05). The architecture described is specific, not hand-waving: onboard GPU running local vision models for live tracking, cloud LLM for semantic understanding, a function-calling layer translating speech into drone actions, 5G link. Demonstrated: voice commands such as "back up 50 meters", camera tilt, target tracking, parking-lot reconnaissance, landing. That source is independent of the company, it predates the cheque, and it is the single best evidence that something real exists.

His GitHub is dark and is not the evidence. maxwelllwang, 23 public repos, all coursework and school-era. No drone, robotics, autopilot, MAVLink or computer-vision repo. Most recent push is a fork of pipecat, an open framework for voice and multimodal conversational agents, on 2026-02-11, which is a natural-language interface tool rather than an autonomy tool. Org dispatchersystems created 2025-06-08, last touched 2026-07-28, 0 public repos, 0 public members, 0 followers (GitHub API, 2026-10-05). Private repos are invisible, so this is absence of signal, not evidence of absence.

One soft spot. The site says "Founding Engineer at Paladin Drones". Paladin was founded in 2018 and went through Y Combinator that year; Maxwell joined in 2020. He was an early engineer, not a founding one. His own LinkedIn carries the title, so this is generous startup convention rather than fabrication, but ask what the scope actually was.

Nick Clouse, co-founder and CEO Harmonic person 194665841

US Naval Academy BS Applied Mathematics, Harvard Business School MBA. Roughly 40. He is not a builder and that is the correct read: he is the domain, customer and defense half. Independent detail on the Anduril role from ZoomInfo (read 2026-10-05): product manager for "perception and machine learning operations for counter intrusion and counter unmanned aerial systems". Counter-UAS perception PM at Anduril is the single most relevant line on either founder's resume for this company, and it pairs with nine years flying fast jets.

The "xAI" claim is verified, not fabricated. The "SpaceXAI" string is a LinkedIn entity-resolution artifact and an independent @x.ai email record corroborates it. The site's "Meta, Anduril, and xAI" is accurate.

His full-time status is unresolved and it is the biggest open item on the team. Harmonic lists four concurrent current positions: Dispatcher founder, xAI Human Data, StickMonkey founder, and a generic "Tech" product manager. ZoomInfo describes him in the present tense at x.ai. The company site lists xAI only as past experience. The LinkedIn profile scraper hit its free-tier run cap, so this could not be settled from outside. Ask him directly. StickMonkey (flystickmonkey.com, Harmonic company 54853727, founded 2024, "aircraft sharing and flight scheduling", headcount 1, no funding) looks like a dormant one-person side project but is still listed as current.

A different and unrelated Nick Clouse exists (SoFi engineer, Montana, Harmonic person 195602918). Do not conflate them. No evidence either founder has operated under another name.

Kyle Mumma, founding engineer, and the best execution signal on this team Harmonic person 2907258

Google SWE intern 2022. University of Washington BS Computer Science 2019 to 2023. Sentry software engineer 2024-01 to 2025-02, then SWE II 2025-02 to 2026-06. Dispatcher from 2026-06. Maxwell Wang was at Sentry 2023 to 2024, so they overlapped at Sentry. A $500,000 pre-seed that can pull a credentialed SWE II out of a 445-person Series E company on a warm, verifiable relationship is recruiting above its weight. He is not on the website, which has not been touched since 2025-10.

Walter Obasuyi, advisor, not staff Harmonic person 111425664, role ADVISER from 2026-06-01

National Guard military police 2011 to 2022. Shelby County Sheriff's Office deputy sheriff 2016 to 2023. Nike security manager. xAI security operations centre lead 2024-12 to 2025-06. SpaceX security protective operations centre supervisor 2025-06 to 2026-09, then SpaceX senior manager security from 2026-09. Herzing MBA. He is a real advisor with a long independent trace and he is the exact buyer persona for the security use case, the opposite of a decorative name on a slide. He came through the founders' own network: he overlapped with Nick Clouse at xAI and with Maxwell Wang at SpaceX.

Caveat. Do not model him as a channel into SpaceX. SpaceX is his employer, so any SpaceX sale runs straight into his conflict of interest.

Charlotte Varne, outbound business development, France unverified

Appears in the LinkedIn roster with the correct company id, located Boulogne-Billancourt. No Harmonic person record found (checked charlotte-varne and charlottevarne). Most plausibly a contract SDR. Do not count her as core team until confirmed.

Adverse sweep: what completed and found nothing

SEC EDGAR full-text on the company and both founders. usaspending.gov spending_by_award for contracts, grants and IDVs. Hacker News Algolia, zero hits for "dispatchersystems". Web search for litigation, fraud, complaint and controversy against the company and both founders by name. Wayback claim-removal check across three homepage captures. All 2026-10-05, all clean.

Prior-company check on Maxwell Wang: Paladin Drones is alive and growing, not a failure he walked away from (Harmonic: $33,998,621 raised over 7 rounds, Series A, 196 headcount, investors including Y Combinator, Khosla Ventures, Thiel Fellowship and Toyota Ventures). Glassdoor was deliberately not attempted; with a real team of three there is no review base.

House patterns checked specifically: inflated traction not found, for the unusual reason that the site makes no traction claim at all. Fabricated team claim mostly clean, one soft spot (the Paladin title). Renamed founder not found. Claims appearing only on the company's own site one, and it is trivial (only the site calls them CTO and CEO; a16z and LinkedIn say "Founder" for each), and the inverse holds, because the site understates the team by omitting Kyle Mumma. Advisor or investor list with no independent trace not found.

The central claim is never defined anywhere public, and that absence is the finding

Across the home page, both legal pages, the a16z profile, both founder bios and the demo day video there is no record found of any mention of PX4, ArduPilot, MAVLink, MAVSDK, DJI SDK, DJI Payload SDK, DJI Mobile SDK, Skydio SDK, ROS, ROS 2, Pixhawk, Auterion, onboard compute, NVIDIA Jetson, QGroundControl, Mission Planner, RTSP, ONVIF, Remote ID or any named airframe. For a product whose entire wedge is integration breadth, publishing zero integration surface after nearly twelve months is unusual. The App Store listing is the only place the company states a hardware scope, and what it states is DJI plus one Parrot variant.

The honest read. "Any drone" is almost certainly either PX4 and ArduPilot via MAVLink, which is not "any", or a video-plus-telemetry overlay that reads streams and issues coarse waypoint commands. The pipecat fork and the "5G or Wi-Fi required for streaming" line are more consistent with the second. The App Store hardware list settles the commercial question regardless: it is DJI.

The four website product "screenshots" are design mockups, demonstrably so

1. A typo in the interface copy. The COMMAND INPUT image reads "1 mile radious around point echo is a trespassing zone." "Radious" is not a word.
2. The system's own responses are blank placeholder bars. Operator messages render fully; every drone reply (DRONE A63, DRONE B21) is an empty rounded rectangle. The exact thing the product is claimed to do is the one element left as a wireframe block.
3. An impossible coordinate. Two images label drone telemetry 34.0500 N, 198.3137 W. Longitude cannot exceed 180 degrees. No such place exists and real telemetry cannot produce it.
4. Canonical placeholder coordinates everywhere else. 34.0522 N, 118.2437 W is the textbook coordinate of downtown Los Angeles. The footer reads 37.7749 N, 122.4194 W, the textbook coordinate of San Francisco.

The "Patrol route/map" image contains no interface at all. It is a stock vertical aerial photograph of an urban intersection with two green crosshairs drawn over it.

The content is still informative about intent. The scenarios are a trespassing perimeter, a patrol tasked "from point alpha starting at TOT 1205Z", and a contact report: "Visual contact, black sedan, parked, 30 meters from Magazine 4a. One male, standing outside driver's side. Awaiting further instruction." TOT is time on target and a magazine is an ammunition store. The vocabulary is military and the imagined customer is a guarded installation. Note also "Awaiting further instruction", which describes a system that stops and asks, not one that decides.

The only public demo is 118 seconds and about 12 seconds of it is product

It is a single fixed stage camera with zero scene cuts across all 118 seconds at ffmpeg thresholds 0.3 and 0.1, so it is a pitch recording, not an edited film. Product footage runs roughly t=18s to t=30s and contains two things.

• t approximately 21s. An oblique view of a suburban house fire that is a three-dimensional reconstruction, not live video: the ground plane terminates at a hard rectangular boundary with raised edges, the signature of a photogrammetry or Gaussian-splat model rendered in a viewer.
• t approximately 27s. Genuine aerial footage of a structure fire with a ladder truck, engine and charged hose lines, over which the Dispatcher overlay is composited in post, reading OPERATOR A / 04:04 / 34.0522 N, 118.2437 W.

That is the same downtown Los Angeles placeholder coordinate as the website mockups. Real footage of a real fire, labelled with a coordinate that is not where the fire was. The interface is painted onto the video, not captured from a running system. The fire footage is plausibly from Paladin Drones, where the CTO was an early engineer, which would make it prior-employer material.

What the evidence can and cannot support

Verified. A two-minute investor pitch with roughly twelve seconds of composited product imagery, a frozen marketing page with four mockups, and a shipped pre-1.0 iOS app that has iterated across five months.

No record found of any of: continuous unedited flight footage of any length; any published benchmark (no task success rate, no intervention rate, no time to detect, no false-alarm rate, no command-to-actuation latency); customer-captured footage; a conference demo, a technical talk after November 2024, or a live demonstration; any public code; any third-party write-up (Hacker News Algolia returns zero hits for "dispatchersystems").

So the autonomy claim is unevaluated, and it should be scored as neither proven nor disproven. A composited overlay on prior footage is consistent with a working system and equally consistent with a slide deck. It has no discriminating power. What tilts the read toward "something real exists" is not the company's own material at all. It is the independent November 2024 AI Tinkerers live demo, plus an app that shipped Flight Replays, a feature that only matters if somebody is actually flying and reviewing footage.

Dispatcher holds no waiver, and nor does any customer it named

No record found for Dispatcher, Dispatcher Inc, Dispatcher Systems, Nick Clouse or Maxwell Wang, on both a raw grep of all 88 saved pages and a parsed search of the company-name and responsible-person columns. No record found for CoveredSix, Covered Six, Triumph, ARSEC or Ecoplanet either, so the one claimed signed customer and the in-process ones are equally unable to fly the advertised mission on the public record.

For calibration, of those records 755 waive 107.31. Leading holders:

The Axon entry is the shape of a real permission to do what Dispatcher advertises.

Do not over-read the absence, and here is the precise limit. The published list is keyed overwhelmingly to individual certificate holders, with only about 9 records carrying a corporate-looking name, and Paladin Drones is also absent despite being an established and active operator, because public safety agencies often fly under a Certificate of Waiver or Authorization or as public aircraft operations rather than under Part 107. That pathway exists for a government customer. It does not exist for a private security contractor guarding a commercial site, which is Dispatcher's named market. For CoveredSix and Triumph the Part 107 route is the only door and none of them have walked through it.

Part 108 is not law. Verified today against the Federal Register itself, not press.

A Federal Register API query restricted to FAA documents mentioning beyond visual line of sight returns, as the most recent BVLOS rulemaking entries, the NPRM of 2025-08-07, the comment reopening of 2026-01-28 and the reopening denial of 2026-02-10. There is no published final rule. The FAA's final rule went to the White House Office of Information and Regulatory Affairs on 2026-07-10 and has not appeared in the Register. Publication is also not effectiveness: a final rule carries an effective date and compliance phase-ins, so operations under Part 108 begin materially later than publication.

And here is the extension that most matters for calibrating timeline risk. Executive Order 14307, "Unleashing American Drone Dominance", published 2025-06-11 at 90 FR 24727 (full text, retrieved 2026-10-05), Section 4(a), directs that the BVLOS proposed rule issue within 30 days and that "a final rule shall be published within 240 days of the date of this order, as appropriate." 240 days from the order's 2025-06-06 signature is roughly 2026-02-01. As of today the final rule is eight months past the deadline set by the President's own order. Anyone pricing this company on "Part 108 lands soon" should price that number, not the trade-press optimism.

One genuine easing the earlier workstreams did not catch: the Section 927 waiver process

Under Section 927 of the FAA Reauthorization Act of 2024 the agency can now grant relief by waiver irrespective of whether a regulation authorises waivers, applying the same safety standard as an exemption but without requiring the petitioner to show public benefit and without the notice-and-comment step. The notice names four eligibility considerations, and two fit this company squarely: "High-Value, Limited Use Case", which the notice illustrates with first responder and search and rescue missions, and "Emerging Use Case", where the FAA cites EO 14307 and says it will consider "whether an iterative authorization process would help incubate innovative safety concepts".

Read it both directions. It means the cage is less absolute than waiver-or-nothing: there is now a faster, quieter, iterative relief path explicitly designed for exactly this kind of novel autonomy, and it exists today rather than after Part 108. It also means the excuse is gone. The relief attaches to the operator who flies and carries the safety case, not to the software vendor, and eighteen months after a company run by a former Navy pilot and a former Paladin engineer began work, nothing has been filed under any pathway. A founder who had internalised the Paladin lesson would be in that docket.

The DJI and NDAA supply problem, from primary sources

2025-12-22. FCC Public Safety and Homeland Security Bureau, Public Notice DA 25-1086, WC Docket 18-89, acting on a National Security Determination received 2025-12-21, added to the Covered List all uncrewed aircraft systems and UAS critical components produced in a foreign country, together with the equipment listed in Section 1709(a)(1) of the FY2025 NDAA. Note the scope: this is categorical, not a DJI ban. The Section 1709 auto-listing of DJI and Autel, triggered because no agency completed the mandated audit inside the statutory year, is the narrower of the two actions.

2026-01-07. Public Notice DA 26-22 (read in full, 2026-10-05, docs.fcc.gov/public/attachments/DA-26-22A1.pdf) records a further Determination from the Department of War, quoted verbatim in the notice:

"The DoW has determined that UAS and UAS critical components included on Defense Contract Management Agency's (DCMA's) Blue UAS list do not currently present unacceptable risks... This determination includes UAS and UAS critical components included on the Blue UAS list now and until January 1, 2027. Additionally, until January 1, 2027, the DoW has determined that UAS and UAS critical components that qualify as 'domestic end products' under the Buy American Standard do not pose unacceptable risks..."

Both carve-outs expire on 2027-01-01, which is 88 days from today.

2026-03-06. Public Notice DA 26-223 records DJI's Petition for Reconsideration and Autel's Application for Review, both filed 2026-01-21, in ET Dockets 26-22 and 26-23, with oppositions due 2026-04-06 and replies 2026-05-11. No record found of a decision on either as of 2026-10-05.

Separately, NDAA Section 848 (FY2020, extended by Section 817 of the FY2023 NDAA) bars the Department of Defense from procuring or operating covered foreign-made UAS, so the Navy on Dispatcher's traction slide cannot fly DJI regardless of anything the FCC does.

The core contradiction, and the company's own App Store listing is the proof

Covered List status blocks new FCC equipment authorisation, so the large installed DJI base in American commercial hands keeps flying and a software layer can address it today, but that base cannot be replenished. It is a melting asset. The domestic alternative is narrow and is itself on an 88-day clock.

And every single airframe Dispatcher names in its own App Store listing is foreign-produced: DJI Matrice, Mavic and Mini are Chinese, and Parrot ANAFI is French, which is still foreign-produced for the purposes of that action. The company built its entire declared hardware list on the one segment of the US fleet regulation has closed to new units, and it did so in an app first shipped 2026-04-03, four months after the Covered List action took effect. They also named ANAFI UKR, the Ukraine military variant, and not ANAFI USA, which is the Blue UAS and NDAA-compliant one. If they were chasing US defense or public safety, ANAFI USA is the obvious integration and it is not the one they chose.

So the breadth claim delivers its value to the half of the market the company cannot sell into on defense terms, and the half it can sell into is flying hardware that is now unreplenishable. Supporting "any drone" is also where the engineering cost sits. The company is paying its largest integration cost to serve a shrinking fleet. A sharper company would pick Blue UAS airframes and say so, accepting a smaller fleet for a defensible, procurement-eligible position. There is no public evidence Dispatcher has made that choice or noticed that it needs to.

Export control, and whether this is one product or two

Likely EAR, not ITAR, and the regime recently loosened. Military UAS designed for reconnaissance, combat or weapons delivery fall under ITAR USML Category VIII. Commercial and dual-use UAS are controlled under the EAR, principally ECCN 9A012 with software under 9D004 (9D004.e covers software specially designed for the operation of 9A012 items). BIS published "Streamlining Export Controls for Drone Exports" on 2026-08-14 (FR document 2026-16628), removing wind-gust tolerance as a control parameter, raising the endurance threshold from 30 minutes to three hours, and leaving UAVs under 9A012.a.1 with under 3 hours endurance and under 300 km range controlled for Anti-Terrorism Column 1 reasons only, with National Security controls removed. As a pure software layer for commercial short-endurance drones doing site security, the licensing burden is light. That is good news and it is real.

The risk is a migration the company's own stated direction invites. If the Navy relationship matures into a CRADA and the software is tailored for military ISR, target detection and tracking, or a military airframe, classification can move toward 9A610 or USML Category VIII. That means DDTC registration, licensed technical data handling, restrictions on foreign-national engineers (relevant to any San Francisco hiring plan), and a compliance function a three-person team does not have. Adjacent and equally unbudgeted: CUI handling, CMMC for defense contract eligibility, and FedRAMP or an ATO for anything in a government environment. No record found of any export-control posture, ITAR registration, CMMC status or security certification in any public material. Unremarkable at pre-seed, but it is a gap for a company putting a Navy seal on a slide.

Five verticals are two products, and the slide knows it

Product A, persistent perimeter security including defense. Flies continuously, at night, over people, needs multiple aircraft per site and needs to go out of sight around a perimeter. Value is reaction latency. Buyer is a guard-services contractor or facility security manager. This is what the mockups depict and what the traction slide sells. Regulatory exposure is maximal: 107.31 plus 107.35 plus 107.39.

Product B, inspection and survey for agriculture, construction and heavy industry. Scheduled sorties producing orthomosaics, volumetrics, crop-stress indices and thermal maps. Buyer is an ops or engineering manager. Value is the analytics pipeline. Regulatory exposure is minimal: a scheduled daylight flight within line of sight over private land is legal today on a bare Part 107 certificate.

Security and defense genuinely are one product, because a perimeter is a perimeter, and the founder pairing is credibly aimed at it. Agriculture and construction are a different company bolted onto the same sentence.

The uncomfortable implication. The regulatory split runs exactly opposite to the commercial one. The vertical Dispatcher can legally serve today with no relief at all, survey and inspection, is the one with the least pull, the most competition and no need for an autonomy agent. The vertical with real pull, persistent security, is entirely gated behind relief that neither the company nor any named customer holds. There is no configuration of this business that is both legal today and interesting today.

Asylon Robotics is the direct competitor and the comparison is unflattering

Harmonic's own description of it: "the only U.S. company delivering fully orchestrated air and ground robotic operations at enterprise scale". That is Dispatcher's exact pitch, executed by a company eleven years older, with $28.9M raised, 79 people, a reported $12M Air Force STRATFI for base perimeter security (DRONELIFE, 2023-11-07), and decisively four 107.31 waivers that let it actually fly the mission. Dispatcher's differentiator would have to be the natural-language agent layer, which is the one thing with no public evidence behind it.

Paladin Drones matters for a different reason. Maxwell Wang was there 2020 to 2023. Paladin is drone-as-first-responder for public safety, a business whose core competence is operating legally inside exactly the cage described above. That is the single most valuable thing on either founder's resume for this company, more relevant than the SpaceX line the website leads with. Whether the lesson was applied is doubtful: a founder who internalised it would be filing waivers in year one, and the FAA database says nobody has.

One, the ceiling is visible and low

FlytBase is the single most important comparable in this file. It has run exactly this play for ten years: hardware-agnostic drone autonomy over third-party airframes and docks, sold to enterprises and security service providers. Harmonic: founded 2016-02, 81 headcount, $125,000 total raised, effectively bootstrapped on revenue. Reported scale, unverified: roughly $10.8M ARR and about 98 people, "400+ programs in 70 countries", customers including Dole, Discovery Land and MSS Security. Pricing is quote-only.

Now read its supported-hardware list: DJI, Hextronics, Heisha and IDIPLOYER (flytbase.com/platform, 2026-10-05). Skydio is not on it. Flock is not on it. BRINC is not on it. Anduril is not on it. The leading drone-agnostic autonomy layer in the world, after a decade, supports DJI and three dock makers.

Two companies arrived at the same supported-hardware list a decade apart, one with 400 programmes behind it and one at version 0.0.6. "Any drone" is not a product decision. It is a description of which SDKs are open, and the answer is DJI's. FlytBase at roughly $10.8M ARR is the best case for the pure software layer, not the median.

Two, the SDKs are closing and the one that is open is Anduril's

Skydio has no SDK programme; its X10D control and telemetry interface is gated through Support, integrations run through the vetted Skydio Extend partner programme, and the enterprise EULA forbids using the offering to build a competing product (skydio.com enterprise terms, 2026-10-05). Flock sells its own NDAA-compliant Alpha airframe into its own camera and dispatch estate. BRINC arrives bundled with Motorola. Each has every incentive to keep the autonomy on the aircraft, because the autonomy is why the aircraft costs what it does.

Meanwhile Anduril runs an open Lattice SDK with REST and gRPC APIs and published developer docs, launched with ten partners, and in 2026 put Lattice for Mission Autonomy onto a third-party Group 5 aircraft and integrated Origin Robotics' BLAZE interceptor. If your wedge is "open layer over everyone's hardware", a $61B company already did that with the customer relationship attached, and Dispatcher's CEO came from there, so he knows it.

Three, the headline feature is now free

As of 2026 natural-language drone tasking is published research and open source: MCP-plus-MAVLink universal LLM drone command interfaces (arXiv 2601.15486), LLM swarm tasking accepted at IEEE WoWMoM 2026 (arXiv 2605.03788), multimodal-LLM vision-language-action drone agents evaluated for commanding, approaching, tracking and searching (arXiv 2609.01404), and working public repos driving PX4 through ROS2.

Of Dispatcher's five headline mechanisms, number one is a paper and a repo and "alerts by SMS" is Twilio. That leaves adaptive search and live decisioning, and there is no public evidence of either working on a real aircraft.

Four, every exit in this category went to the system of record at a feature price

Airware is the definitive death and the quote is worth having verbatim. It raised $103,934,148 (Harmonic, OUT_OF_BUSINESS) and shut on 2018-09-14. TechCrunch's post-mortem: "Over time, the software that shipped with commercial drone hardware from other manufacturers was good enough to make Airware irrelevant" (techcrunch.com). Airware was an a16z company. That is the exact mechanism that kills this: not a competitor out-executing them, but DJI's and Skydio's bundled software becoming good enough.

PrecisionHawk raised $136,000,000 and filed Chapter 7 in December 2023. Airobotics burned $123,000,000 proving drone-in-a-box hardware and sold into a small-cap public shell. Kespry absorbed into a survey services firm on $61,300,000. Nightingale Security coined the phrase "Robotic Aerial Security", and twelve years in its most recent recorded financing is a $50,000 convertible note from a penny-stock robotics holding company (2024-06-03). That is the base rate in this category, not Skydio.

The ones that survived stopped being software: Auterion sells Skynode silicon, ModalAI sells VOXL boards, Percepto sells the dock and the aircraft. The ones absorbed went to whoever owned the workflow: DroneSense and Dedrone to Axon, Cape to Motorola, Site Scan to Esri, DroneDeploy to Procore.

The Auterion lesson, because it is the main precedent for open middleware

The open commercial drone OS business is gone. What Auterion is today, per Harmonic: HQ Arlington Virginia rather than Zurich, 190 headcount, $192,200,000 raised, last round $105,000,000 on 2025-09-25, and the investor list includes the United States Department of War. Web reporting puts the round at $130M led by Bessemer, of which $25M was non-dilutive capital from the Office of Strategic Capital (The Robot Report, 2025-09-23); Harmonic records $105M equity, so report both and mark the $130M unverified. It sells Skynode, a physical onboard compute module. Partners are Rheinmetall and Lockheed Martin, not Turner Construction.

The thesis survived only after two changes Dispatcher cannot copy. It became silicon, which restores pricing power and makes integration physical rather than permissioned. And it started selling to governments at war, which are the only buyers that pay for openness as a requirement because they refuse to be locked to one airframe supplier. No commercial site-security buyer does that.

Which market is Dispatcher actually in, and who signs

It markets autonomous site security. It ships, if anything, a DJI mission-control app. The real product today sits in a fourth market the website never names, drone flight-ops software for DJI fleet operators, which is FlytBase's market. That market is smaller, already consolidated, and the one where "works with any drone" is simultaneously most true and least valuable.

Counter-UAS, which the founders' backgrounds might suggest, is not what it sells. Counter-UAS is detect, track, identify and mitigate hostile drones; Dispatcher is the opposite side of the problem, tasking friendly ones. That is a plausible future pivot, not a current market, and it would be a pivot into Anduril's and Auterion's teeth.

Who signs for site security. Champion is the facility or site security manager; signature is the corporate Director or VP of Physical Security, with Operations, Legal and Risk in the loop, and the CSO plus Procurement for multi-site. There is almost always a fourth party: the contract guard provider (Allied Universal, Securitas, GardaWorld) who staffs the site today and is either the channel or the opposition. This is why Asylon, Percepto and FlytBase all sell through or to security service providers rather than around them.

What budget. Not IT and not innovation. It comes out of the contract guard line, which is why the pitch is always guard displacement. 2026 anchors, web, unverified: an unarmed 12-hour guard post runs roughly $8,000 to $12,800 a month at about $22 to $35 an hour; true 24/7 coverage of one post runs roughly $16,000 to $25,600 a month unarmed, so $190,000 to $300,000 per post per year, with the guard industry at 77% annual turnover in 2024. That is the number the drone has to beat, and it is why the category exists.

Sales cycle. Site survey, airspace and FAA authorisation review, network and power work at the dock location, a paid pilot of one to three months on one site, union and works-council consultation where guards are organised, insurance and liability review, privacy and employee-monitoring review, then a security-committee or capital gate. Six to eighteen months from first meeting to a multi-site contract. Nothing about that is shortened by a better tasking interface, and all of it is handled today by vendors who arrive with the aircraft, the dock, the waiver and the insurance certificate.

And the construction channel just closed. Procore closed its $845M all-cash acquisition of DroneDeploy on 2026-09-09, pitched explicitly as "see, understand and act on the jobsite", which is Dispatcher's pitch with the system of record attached. Construction is also the worst buyer shape in enterprise software: funded from the project general-conditions budget, charged to the job, so the sale is per-project and re-sold every job. Recurring effort, non-recurring revenue. Agriculture pays the least, has the shortest season and has killed more drone companies than any other vertical (PrecisionHawk, AgEagle, Delair). Treat the agriculture and construction panels as TAM decoration. If Dispatcher pursues them seriously that is a negative signal, not a positive one.

Where the pull actually is, per their own slide. Of the eight logos, six are security or defense, one is a bitcoin miner evaluating site security, and exactly one is agriculture, sitting in the weakest column. The market is telling them what they are and the website has not been updated in eleven and a half months.

No customer anywhere is confirmed by anyone other than Dispatcher

The 122,861-byte homepage was parsed in full: no logo wall, no "trusted by", no case study, no testimonial, no customer name in any heading, body text, image filename or alt attribute. Every image is a generic sector photo (Construction.jpg, Agriculture.jpg, Energy & Heavy Industry.jpg) or a product mock. The only two faces are the founders. There was nothing to verify at the customer end. The finding is the absence.

What the company itself claimed at demo day, October 2025

Nothing has been added to any public surface in the twelve months since. The named set is overwhelmingly contract physical security, which is the real market, not agriculture and not defense.

Federal money: zero, and the method was validated rather than assumed

The SBIR result is now a clean negative and should be read as one. Dispatcher Inc has never received an SBIR or STTR award.

One question for the founders. DIU is running a Voice-Controlled AI Drone Swarm Prize with a reported $100M pool for plain-English command of drone swarms by a single operator, which is close to a verbatim description of Dispatcher's pitch. There is no evidence the company has entered it. If they have not, ask why.

A CRADA is a cooperative research agreement, not a procurement contract, so its absence from usaspending is expected and is not evidence against it. The Navy seal on an "in process" column is a sales conversation.

Measurable usage signals, none of which is revenue

No Android app. No careers page, no /jobs link, no job postings found. No Discord, Slack, forum, newsletter or YouTube channel. Apify returned an empty array for company LinkedIn posts, consistent with a page that has never posted.

Read the app both ways, because it is the one thing cutting in the company's favour. Positive: real shipped software from a real corporate seller, iterated across five months, with concrete feature work ("Flight Replays", "Guest links") that only matters if somebody is flying and reviewing footage, plus operator-grade detail like a Command Center UI tuned for a 13-inch iPad Pro. That is not vapour. Negative: it is pre-1.0 and two ratings is indistinguishable from the founders and their friends.

Runway, and why there is a clock. On $500,000 received 2025-07-28 against a three-person San Francisco burn of roughly $25,000 to $45,000 a month fully loaded, runway is eleven to twenty months, so exhaustion lands between mid-2026 and early 2027. Harmonic currently_raising is null, which is neither confirmation nor denial. Combined with the a16z follow-on window closing around 2027-01-28, this company needs a priced round inside roughly four months and has no commercial proof to price it with.

Arithmetic

The adjustment and why. Plus 0.25, the maximum allowed, applied for one reason: the mean is dragged by traction at 1.5 and market structure at 3.0, both correctly scored, but a simple mean under-weights the factor that most predicts a pre-seed outcome, which is the team, and it under-weights the only cheap falsifiable test available, which is whether code ships. An app that went from nothing on 2026-04-03 to v0.0.6 on 2026-08-29 with Flight Replays is evidence that three people are building rather than fundraising.

A reader who declines the adjustment lands on 3.42, and COMPANY 5.42 instead of 5.67. That changes nothing about the decision. Both sit in the same band, both produce the same verdict, and nobody should argue about the quarter point.

EV bonus, +1.00

Harmonic records $500,000 total raised, one round, 2025-07-28, ACCELERATOR_INCUBATOR, sole investor a16z speedrun. Money verifiably received only. That is the $0 to $1M tier. No credible evidence of unrecorded capital was found by any workstream, and the hunt was run properly (no Form D on EDGAR full-text or browse-edgar, empty funding_rounds, null currently_raising, no press figure to contradict).

Two sensitivities so nobody has to re-derive them. First, the a16z follow-on half is a right and not cash, and even if drawn in full the cumulative $1.0M stays inside the $0 to $1M tier under the boundary rule, so the EV bonus is insensitive to the Speedrun follow-on. Second, only a priced round above $1M moves it: a $3M seed takes EV to +0.75 and COMPANY to 5.42, a $5M seed takes EV to +0.50 and COMPANY to 5.17. Nothing found today moves the tier.

Geo bonus, +1.00 (USA)

Harmonic gives HQ San Francisco, California, United States, metro San Francisco Bay Area. a16z's own portfolio page renders "Located in San Francisco, California". The operating entity of record is Dispatcher Inc., corroborated by two artifacts independent of the company's own marketing: the Apple App Store seller of record is "Dispatcher Inc." with seller URL dispatchersystems.com, and the site footer reads "DISPATCHER INC." Harmonic legal_name is null.

State of incorporation is unverified. California Secretary of State bizfileonline.sos.ca.gov returned HTTP 429 again today, naming our droplet IP 159.203.137.60 with the body "This request was blocked by our security service", and Delaware is captcha-walled. Rule applied: USA +1.00 on verified US operating presence and a verified US seller of record, with the registry confirmation flagged as an open item rather than claimed.

Chain dock, 0.00

Not a crypto company. No token, no chain, nothing on Berachain. The only crypto-adjacent item anywhere is MARA Holdings appearing as a prospective perimeter-security customer, which is a buyer and not a chain dependency.

Is there a round? No, not yet, and that is the opportunity and the problem.

Harmonic currently_raising is null, there is no Form D, and nothing has been announced. But the constraint is hard: $500,000 received 2025-07-28 against a three-person San Francisco burn means runway is thin to gone, and the a16z follow-on $500,000 requires a third party to lead and price a round inside a window closing around 2027-01-28. So a financing event is effectively forced inside four months, and a $100,000 to $500,000 cheque is a meaningful part of a round this size. Daxos could actually matter here, which is not true of most things in this sector.

What a Speedrun company's next round prices at, and who gets allocation

Speedrun writes one non-negotiable $500K-for-10% SAFE to every accepted company, takes pro rata, and takes no board seat. That matters for reading the signal: every company in a 60-plus-company cohort gets identical terms, so the cheque says nothing about how a16z ranks Dispatcher inside SR005, and it is not an a16z bet in the way a Series A from a16z proper would be. Treat SR005 as a credential on the founders, selected at a reported sub-0.5% acceptance rate from more than 14,000 applications, and nothing more. Worth noting too that SR005's published brief was games, tools, platforms, XR and next-generation entertainment, which is a consumer-leaning call for a defense drone company to be admitted under.

On pricing, the cohort splits sharply. A Speedrun graduate with real traction raises a seed in the low-to-mid teens post-money and the allocation is gone, consumed by the lead plus a16z's pro rata, with nothing left for a $100K to $500K angel. A Speedrun graduate with no named customer, no revenue, no waiver filed and a frozen website raises a bridge or a small priced round at the bottom of the range, and there is plenty of room for warm money because there is no competition for it.

Dispatcher is unambiguously in the second group. Daxos will not be squeezed out of this one, and the reason Daxos will not be squeezed out is itself the finding.

What Daxos can and cannot add

Cannot add the two things that bind: FAA relief and SDK access. Those are earned by the operator who flies and carries the safety case, and by vendor cooperation that incumbents are commercially motivated to withhold.

Can add real value on the one axis where the pull actually is: introductions into contract physical security and facility security management, which is where six of the eight logos on their own slide sit, plus defense channel relationships. That is a genuine contribution and it is unusually well matched to this company's revealed market rather than its advertised one.

Why 4.5 and not higher or lower. Higher would require one verified customer or one filed waiver, either of which would move this immediately. Lower would ignore that the team is in the house's lane, is checkable and checked out, and that a forced financing event in four months gives Daxos a real seat.

VERDICT: MEET

MEET means take a diligence call, not an intent to invest. The case for meeting rather than watching is specific: there is a forced event inside four months, Daxos has a real seat in a round this size, and the three artifacts that would settle the entire thesis are cheap for a working system to produce and impossible for a non-working one.

The case against going straight to IN is equally specific: nothing commercial is proven, the declared hardware list is on the wrong side of a regulation that took effect four months before the app shipped, and the autonomy claim has no independent demonstration.

The default outcome of that meeting is PASS. It becomes something else only if the founders produce what is listed below.

Meet now, on the clock. Convert to IN only if two or more of the following are produced, with at least one from the first three.

1. One continuous unedited recording of the viewer's own novel natural-language task, aircraft and screen both in frame, no cuts, no post-production overlay. The single highest-information artifact, and it costs a working system one afternoon.
2. The intervention rate per flight hour over the last 30 days, plus the autopilot and command interface named explicitly. PX4 and MAVLink, DJI PSDK, Skydio Extend, or a ground-station overlay. "Any drone" is not an answer.
3. One nameable site where Dispatcher has flown a paid or unpaid patrol, with the operator's Part 107 certificate and the specific relief it flew under, confirmable with that operator rather than with Dispatcher.
4. An FAA filing under Dispatcher's own name: a 107.31 or 107.35 waiver application, a Part 11 exemption petition, or a Section 927 waiver request under Docket FAA-2026-1100. Filed, with a docket number. Not "planned".
5. One NDAA-compliant or Blue UAS listed airframe in the supported-hardware string. The strongest single available signal, because it implies real SDK access rather than DJI MSDK and it fixes the procurement contradiction.
6. A priced seed with a named lead, which would also tell us whether a16z follows its own Speedrun company.
7. Nick Clouse confirmed full time, with the xAI role ended.

Reopen automatically on either external trigger, regardless of anything the company does: Part 108 published in the Federal Register with an effective date, or a renewal or lapse of the Blue UAS and Buy-American carve-outs on 2027-01-01. Both change the shape of this business more than the company can.

One free tracking instrument. The App Store listing is a public, dated, versioned telemetry feed. Re-pull itunes.apple.com/lookup?id=6761081577 in 60 and 120 days. If the version climbs and a dock integration or an NDAA-compliant airframe appears in the supported-hardware string, the thesis improves materially. If it still reads v0.0.6 with two ratings, the company is stalled and this file stands.

1. Name the autopilot and the command interface, then name one site where Dispatcher has flown a patrol and the certificate and relief it flew under. Neither you nor CoveredSix nor Triumph nor ARSEC appears anywhere in the FAA's 1,595 Part 107 waiver records, so if the answer is "within line of sight with a pilot standing there", what is the software worth at $50,000 a site?
2. Nick, are you full time at Dispatcher, and have you ended the xAI role? LinkedIn lists xAI Human Data as current alongside StickMonkey and a generic product-manager role, and ZoomInfo still carries an @x.ai address for you. This is the one thing about this company that could not be resolved from outside.
3. Every airframe in your own App Store listing is foreign-produced and inside the FCC's 2025-12-22 Covered List action, and the Blue UAS and Buy-American carve-outs lapse on 2027-01-01. You shipped that list in April 2026, four months after the action. Which NDAA-compliant airframe is on the roadmap, why ANAFI UKR rather than ANAFI USA, and how do you sell to the Navy on your traction slide when Section 848 bars it from flying anything on that list?

Named plainly so none of these is read as a clean negative.

• Litigation. No check was performed on the company or on either founder. CourtListener returned "Request was throttled. Rate limit exceeded: 125/day" on every query, and I reproduced the HTTP 429 again today across three separate searches. The v3 endpoint rejects anonymous users. Needs an API token or a reset window. A genuine gap.
• State registry. California SOS returned HTTP 429 with "This request was blocked by our security service" and our droplet IP named, reproduced again today. Delaware is captcha-walled. State of incorporation and entity standing are unverified, and neither founder was checked for dissolved or delinquent prior entities. The name "Dispatcher Inc." is corroborated three ways (Harmonic location, App Store seller of record, site footer) but the registration itself is unconfirmed.
• Whether any code exists. The GitHub org has 0 public repos. Private repos are invisible. Absence of public code is not evidence of absence of code.
• Which SDK or autopilot is integrated. No public disclosure anywhere, on any surface, after twelve months.
• The Navy and CRADA claims. A CRADA is not a procurement contract, so its absence from usaspending is expected. The SBIR negative is now definitive but does not speak to a CRADA.
• Whether DJI's Petition for Reconsideration and Autel's Application for Review have been decided. ET Dockets 26-22 and 26-23, both filed 2026-01-21. No record found of a ruling as of 2026-10-05. A reversal would change the supply analysis materially.
• Reddit sentiment. The search JSON endpoint returns non-JSON from this droplet. Channel unchecked.
• FAA Part 48 small-UAS registrations and the COA holder list. Neither is published. Structurally unavailable, so there is no public way to tie airframes to this company or to find a COA it might hold.
• The actual a16z Speedrun SAFE terms for this company. The $500K-for-10% structure is Speedrun's published standard, not a document anyone has seen for Dispatcher. The implied roughly $5M post follows from the published standard, not a term sheet.
• Whether the demo-day traction claims were ever real. CoveredSix, Triumph and ARSEC were not contacted. Doing so is the obvious next step and it does not need the founders' permission.

House anchors in the adjacent lane, for placement only: Apyx 9.0, Askari Defense 8.5, BESXAR 7.5, Sophia Space 7.5, Archaius 7.5, Satlyt 6.75, Space DOTS 3.67, Auron Solutions 2.5. Thermopylae Aerospace and Aethero sit on the BORO-SM list where all partners voted in.

Dispatcher at 5.67 places below every frontier-hardware name the house has rated as investable and above the two it effectively killed. That is the right neighbourhood and the reason is structural rather than sentimental. Askari and BESXAR carry something Dispatcher does not: a built artifact with a demonstrable performance envelope, or a procurement relationship with a name attached. Satlyt at 6.75 is the closest shape, an early software-and-compute play in a regulated hardware lane, and it sits above Dispatcher because the thing it does is specified and its customer set is nameable. Space DOTS at 3.67 and Auron at 2.5 are below because their claims failed verification; Dispatcher's claims are unverified, which is a materially different and better position than falsified, and that difference is most of the gap between 5.67 and the bottom of the distribution.

Within its own market rather than the house's, the comparison that actually prices this company is Asylon Robotics at $28.9M, 79 people, eleven years and four 107.31 waivers, and FlytBase at ten years, roughly $10.8M of reported ARR and a supported-hardware list that is still DJI plus three dock makers. One shows what it costs to be allowed to fly the mission. The other shows the ceiling on the pure software layer. Dispatcher is three people and $500,000 against both.