Combined Field Intelligence - 2 sources

UAV Discovery: Dimas + Serhii

Two Ukrainian units, two interviews, merged - what they agree on, where they differ, and what is still open

Dimas (Dima) DIMAS

UAV commander, 155th Brigade · call 1 (technical) early; call 2 (logistics) ~2026-05-25, 4m26s · written specs 2026-05-28

Serhii Smirnov SERHII

419th Separate UAV Battalion (drone-only), high-level / defines targets · smirnovsergii@gmail.com · video call Fri 2026-05-22 13:00 CET · intro by Sayed

What this is

One merged intelligence sheet from both UAV interviews. Every claim is tagged by who said it: DIMAS SERHII or BOTH when the two independently agree. They are different units, so agreement between them is strong signal, and disagreement tells you the answer depends on context.

Picture this: Monday, you and AJ are deciding what to actually prototype. Instead of re-listening to two long recordings, you open this. The green "BOTH" cards are your safest bets (two separate front-line units said the same thing). The divergence table stops you from designing for one unit's reality and breaking the other's.

1. Where both men agree HIGH CONFIDENCE

Two different units, same conclusion. Treat these as the load-bearing facts for any prototype decision.

~20% hit rate is real

About 1 in 5 drones reaches the target, and it is roughly independent of which drone you use.

Dimas: 4-6 of 10, sometimes 2.
Serhii / Sayed: 1-2 of 10; "up to 20% regardless of what drones you use."

Autonomous aiming has already failed

Both units tried or saw AI / auto-aim drones and dropped them. This is the single most important thing to internalize before pitching AI.

Dimas: another brigade tried AI drones, "fuck this, doesn't work", went manual.
Serhii: tested a couple - 3 to 5 min to lock, target "didn't catch", imprecise even when locked.

Friendly fire kills the AI-targeting idea

Neither will let a drone autonomously pick a human target near their own troops. The manual operator stays in control by design.

Dimas: non-AI drones already sometimes hit own soldiers; AI could hit one plus a second nearby.
Serhii: in direct contact (enemy 30-50 m from own men) they switch to fiber or heavy bombers, not signal FPVs.

Trees, forest and concrete break everything

The hardest environment for both signal and targeting. Any prototype has to assume obstructed line of sight, not open field.

Dimas: targets hidden in tree lines are hard to find; AI will not make it easier.
Serhii: signal lost through trees / forest / concrete; main natural cause of loss.

Fiber-optic is expensive

Both flag cost as a real constraint on fiber, even though it dodges jamming.

Dimas: fiber-optic drone roughly 75,000 UAH (~$1,800-2,000), not $75k USD; Starlink/satellite drones cost more, which is the main barrier for satellite.
Serhii: the fiber itself is "really costly"; by late May he calls fiber "not really in demand right now because of its cost and cost efficiency"; long-range Bulava complex even more so.

Both can buy, test and give feedback

Both have a real path to procurement and offered honest field feedback on prototypes - via others, not personally.

Dimas: buys with brigade budget; has specialists + a training camp + a front line.
Serhii: govt budget + can reach commanders/decision-makers; wants a 10-12 unit batch for real statistics.

2. Where they diverge

Different units, different doctrine. Design for both or you will fix one unit's problem while ignoring the other's.

Topic
Dimas (brigade)
Serhii (unmanned battalion)
Top pain point
Fiber-optic cable quality + battery quality. Bad components stop drones reaching targets.
Signal loss / jamming first, then targeting speed and precision. "The most important thing is the signal."
View on fiber
Prefers fiber - it beats radio jamming, which freq-hopping cannot fix.
Regular FPV still best for hitting; fiber is slower to launch, heavier, sensitive, gets cut by obstacles/wind. Uses fiber for direct-contact + recon.
Cable failure mode
Cable is simply "low quality" (no mechanism given).
Concrete: fiber gets cut by trees/obstacles, lost to strong wind, heavier + more sensitive, slow setup.

3. Full side-by-side

Everything each source said, topic by topic. Blank / not raised means that person did not cover it.

TopicDimas DSerhii S
Unit Regular brigade; commander officer. Separate unmanned (drone-only) battalion; high-level role, defines targets, 2 years in.
Drone types Fiber-optic (preferred), radio + digital, Starlink/satellite. Fiber carries only video + control; the drone still flies on its own onboard battery. FPV (manual), heavy bombers (Baba Yaga, at night), Mavic recon, fiber-optic, DJI Matrice retranslators (relays), Bulava long-range complex.
Named models Goryn (top digital camera, but landing/release unreliable) · Uryi (balanced, good load, unstable in strong wind) · Ronni (stronger optics, worse camera) · Beshket ("top of the top"). All fiber-optic, payload under 4.5 kg. Fiber: Vyriy, Gromylo, Beshket (mostly 10/13"). Radio: Heneral Chereshnya, F10, Rarog, Vyriy, Shrike, Kalibri (mostly 7/8/10/13"). From the Brave1/Delta catalog.
Anti-drone (counter-UAS) "Very useful and relevant now, they do a lot of work on the front." Best per Dimas: DROZD and GENERAL CHERESHNYA. Not raised in this interview.
Payload / range Payload 1.5-4.5 kg; fiber operating range 10-30 km by target distance. Fiber range 15-20 km for zone circles, 25-30 km to locate bases/staging/equipment.
Preferred to hit Fiber-optic (anti-jam). Regular FPV (fiber too slow/costly); fiber only for direct contact + recon.
Hit rate 4-6 / 10, sometimes 2. 1-2 / 10 good conditions; ~20% regardless of drone.
Jamming Radio jamming severe; frequency-hopping does NOT help, so fiber. Loses more to Russian jamming. Tactical jammers (10-50 km) avoidable via mid-flight channel switching; deep-area (>20 km, near cities) jammers defeat them.
Auto-aim / AI Heard it "doesn't work", units go manual; skeptical for humans, maybe OK for vehicles / large targets. A previously-contacted brigade tried AI drones, said "fuck this, it doesn't work" and reverted to manual. Tested a couple: 3-5 min lock, imprecise. Only works at 10-50 m from target, not 500 m. Operator must still see target. Root-cause guess: not the processor but camera quality + software quality (Sayed proposed software optimized to run faster on the same processor).
Long-range strike Not raised. Bulava complex flies over 60 km: a wing-type secondary targeting/surveillance drone ("Swavans") loiters high (~500-1000 m), detects and locks the target, then a connected strike drone is released and hits (laser-guided-missile style). Needs two operators (one to lock, one to release). Very costly. Used vs groups, heavy equipment, important buildings.
Signal loss cause Not detailed. Radio is line-of-sight; lost low (5-10 m) or higher; trees/concrete; retranslators reach 400-500 m but fail if enemy near trees and relay 10-15 km off.
Crew per drone Not specified (manual FPV operator). ~3 per position: 2 aim/wind + 1 spotter (arms + places drone).
Daily volume Not specified. 10-15 drones/day per position (quiet days); ~10 fiber flights/day.
Training time Not specified. ~1 month + co-pilot field time for a new FPV operator.
Cost Fiber drone ~75,000 UAH (~$1,800-2,000, NOT $75k USD). Satellite/Starlink drones cost more than fiber-optic, which is their main barrier. Fiber "really costly" and, by late May, "not really in demand right now because of its cost"; Bulava complex very costly.
Procurement Buys directly with brigade budget if a prototype proves useful. Govt-supplied + self-purchase budget (approved allow-list only); production needs Brave1 listing/certification (donations exempt; certification reportedly ~24 hrs now, far easier than before the war); plus "Yabali" e-points gamification (verified kills to points, redeemed on Brave1 market).
Test path Specialists trial in training camp, then field, then feedback. Passes to commanders; needs ~10-12 unit batch for clear statistics; can connect decision-makers.
One thing to fix Not framed this way (implied: component quality). "The signal" - signal loss above all.

4. Your open questions - now cross-checked against both

Serhii answered things Dimas did not. Re-scored so you only ask what is genuinely still missing.

QuestionStatusWhat we now have
Km range strike drones fly today PARTIAL Serhii gives surrounding numbers (retranslators 400-500 m altitude, tactical zone 10-50 km, Bulava 60 km+) but no clean fiber/FPV strike range. Still worth one crisp number from each.
Km where cable quality degrades STILL OPEN Neither gave a distance threshold. Ask Dimas specifically (it is his pain point).
Cable problem symptoms ANSWERED Dimas: on cable break both signal and control are lost and the drone simply disappears; failure distance is manufacturer-dependent (some unusable from the start, some snap mid-flight, two batches from the same maker can differ). Serhii adds: cut by trees/obstacles, lost to strong wind, heavier + more sensitive, slow setup; the longer the flight on the tether, the higher the snap risk.
Payload envelope (1.5-4.5 kg) ANSWERED Dimas: payload 1.5-4.5 kg; his named models fly fiber-optic under 4.5 kg. Still worth confirming whether the figure is payload carried besides the airframe or all-up.
Drone specifics / brands / website PARTIAL We now have model names: Dimas runs Goryn / Uryi / Ronni / Beshket; Serhii lists Vyriy / Gromylo / Beshket (fiber) and Heneral Chereshnya / F10 / Rarog / Vyriy / Shrike / Kalibri (radio) from the Brave1 catalog. Still missing: the manufacturer websites / detailed specs behind Dimas's four named models (requested, not yet provided).

5. What this means for the prototype + next questions

The convergent, low-risk bet is NOT autonomous human targeting. Both units have already tried AI/auto-aim and rejected it, and both fear friendly fire. Pitching "AI that picks and hits humans" walks straight into their strongest objection.

The opening both of them actually point at is kill-chain reliability:

  1. Signal resilience / anti-jam / low-latency link - Serhii's #1 problem and the cause of most of the missing 80%. An AI/IoT angle here (smarter relay handoff, channel selection, link recovery) lands without the friendly-fire fight.
  2. Fast terminal lock in the last 10-50 m, not 500 m - that is exactly where the tested systems failed (3-5 min, imprecise). Assist the operator who already has eyes on target; do not replace him.
  3. Component quality - Dimas's fiber cable + battery pain. Less glamorous, possibly the fastest measurable win.
  4. Frame AI as operator-assist, manual control retained - both insisted the human stays in control. Make that the headline, not the footnote.

Next call, keep it to four light asks (do not bombard): 1) clean strike-drone range, 2) the 4.5 kg payload definition, 3) specific FPV/fiber models or a website, 4) for Dimas only, the distance where cable quality starts failing. Everything else both men already answered.