This catalogue measures how technical demo, launch and explainer videos for developer tools present code. The files hold no view, click or sales data and no audience field, so visually present means the treatments the coding can see, whether one works is the reader's test, and engineering buyers means engineers as the buyers of these products, judged from product category and script. 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026, qualified: a thin set, under this series' floor of 80. A voiceover carries 42 of 49, 33 of 49 run 1:30 or less, and 29 of 42 scripts address the viewer as you at least twice per 100 words. Among the 13 with full visual coding, screen capture builds 8 and code reaches the screen in 8, 4 of them from Anthropic and OpenAI. GitLab, Pinecone, Microsoft and Netlify are in the set.
945 videos were screened against a rule written before screening; 49 qualified, from 43 companies. Every one is a demo, launch or explainer video for a developer tool or platform sold to engineers, names code or a developer environment in its title, description or script, runs 0:25 to 2:45 and was uploaded to YouTube between 2 January 2024 and 3 August 2026, 32 of them in 2026.
ADVIDS Research · 49 developer-tool videos coded · 43 companies · 42 scripts counted · 13 visually coded · published 14 September 2026 · 18 min read
Top takeaways
Screen capture builds 8 of 13 visually coded developer-tool videos
8 of 13 visually coded developer-tool videos uploaded 2024 to 2026 put a screen capture in the build and 4 an interface mockup; the 5 without a capture: Anthropic, GitLab, HPE, LangChain, etc.
8 of 13 put code on screen; 4 are Anthropic or OpenAI
8 of 13 visually coded developer-tool videos put a terminal, code text or screen capture on screen, 4 of them from Anthropic and OpenAI; the 5 without code on screen: Airia, Akto, Cequence Security, HPE, etc.
A voiceover carries 42 of 49 developer-tool videos
42 of 49 developer-tool demo, launch and explainer videos carry a voiceover; the 7 on music alone: Anthropic, CData, LangChain, OpenAI, etc.
29 of 42 scripts talk to you; 13 name developers
29 of 42 developer-tool scripts say you at least twice per 100 words, while 13 name developers or engineers: Microsoft, Acurast, CyberArk, Deepgram, etc.
27 of 42 scripts use under three developer terms per 100 words
27 of 42 developer-tool scripts use fewer than three developer terms per 100 words, a median of 2.6; Microsoft Agent Framework runs at 13.04.
33 of 49 developer-tool videos run 1:30 or less
33 of 49 developer-tool demo videos run 1:30 or less, a median of 1:13; the 8 over 2:00: BrowserStack, Cequence Security, Modulate, Oracle, etc.
29 of 49 run light or mixed frames, not dark
29 of 49 developer-tool videos run light or mixed frames and 20 dark, yet code sits on a dark ground in 5 of the 6 whose frames show code; LangChain puts its evaluator code on a light ground.
Questions this report answers
How do developer tool demo videos show code in 2026?
Among 13 visually coded developer-tool videos uploaded 2024 to 2026, 8 show a terminal, code text or screen capture, and Anthropic and OpenAI supply 4 of those 8. Pinecone and Cequence Security show no code on screen.
Should a developer tool video use a screen recording or an animated mockup?
In this set the capture leads: 7 of 13 visually coded videos use a screen capture alone, 3 a mockup alone, 1 both and 2 neither. Among the 8 with code on screen, 5 are captured and 3 drawn; Anthropic records Claude Code, LangChain draws LangSmith.
Do developer tool videos use dark mode?
20 of 49 developer-tool videos uploaded 2024 to 2026 run dark frames, 23 light and 6 mixed. Where cut frames show code, 5 of 6 put it on a dark ground, 3 of them inside light or mixed frames, GitLab among them.
Do developer tool demo videos use a voiceover?
Mostly: 42 of 49 are voiced, and speech covers at least 60 per cent of runtime in 38. The 7 on music alone include all four Anthropic and OpenAI videos in the set: Claude Code twice, Claude Managed Agents and Codex.
How technical should a developer tool video script be?
The median script carries 2.6 developer terms per 100 words and 3 distinct terms, and 2 of 42 name a programming language or git. Microsoft Agent Framework runs at 13.04 developer terms per 100 words; Akto at 0.94.
How long is a developer tool demo video?
The median of 49 developer-tool demo, launch and explainer videos runs 1:13, with the middle half between 0:55 and 1:42. Superblocks runs 2:25; Codex from OpenAI runs 0:46.
How was this analysis done?
ADVIDS Research screened 945 videos against a rule written before screening; 49 from 43 companies selling developer tools qualified, under this series' floor of 80. Scripts were counted against a fixed developer-term lexicon, and 13 videos carry frame coding.
How to read this report
The set is 49 developer-tool demo, launch and explainer videos; a script is a caption track of 15 words or more, and 13 videos carry full frame coding. A figure such as 8 of 13 means 8 videos out of a base of 13, and every count names its base: all 49, the 42 scripts, the 13 visually coded videos, the 8 of those with code on screen, or the 6 whose cut frames show code. A pattern holds for at least half its base and at least 8 videos; where removing its largest company moves the share 10 points or more, the header names who supplies it. Bases under 20 are written as counts, and no periods are compared.
Pattern 01 · Build · Screen capture · 13 visually coded videos
1. Screen capture builds 8 of 13 visually coded developer-tool videos
8 of 13 visually coded developer-tool videos uploaded between 2 January 2024 and 3 August 2026 list screen capture in their build, a screencast of the product, against 4 that list an interface mockup. 7 use capture alone, 3 a mockup alone, 1 both and 2 neither. Among the 8 that put code on screen, 5 are captured and 3 drawn. Build codes are medium confidence for 2 of the 8. Anthropic records Claude Code itself, diff and terminal included. The product reaches the screen as a recording twice as often as a redraw.
Companies in this group include Anthropic, Airia, Akto, Cequence Security, etc.
LangChain draws the fix instead of recording it. Its May 2026 LangSmith Engine video runs 1:12 of interface mockups and 2D animation on light panels: an evaluator written as a JavaScript function, a proposed diff to a system prompt and a Merge button, with no screen capture in the build and music in place of a voiceover. Where the product is an assistant that proposes code changes, a drawn diff shows the change without a live account.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
A video picks one way to put the product on screen: capture and mockup share a build once, in ScienceLogic.
Pattern 02 · Code on screen · Code on screen · 13 visually coded videos
2. 8 of 13 put code on screen; 4 are Anthropic or OpenAI
8 of 13 visually coded developer-tool videos uploaded between 2 January 2024 and 3 August 2026 put a code environment on screen: a terminal segment in 4, confirmed code text such as a file path, an import line or an API call in 5, and a screen-capture segment in 3. 3 of the 8 are Anthropic videos and one is from OpenAI; without Anthropic the count is 5 of 10, and without both 4 of 9. Where code is on screen it covers a median 16 per cent of runtime and first appears at a median 0:09; it covers more than a quarter of runtime in 3, all from Anthropic. Frames cut at the code segments show code in 6 of the 8. GitLab opens by typing an import line. Three of the four terminal videos are Anthropic launches.
Kinetic type or a diagram sits in the same video as code in 4 of the 8.
Companies in this group include Anthropic, GitLab, LangChain, OpenAI, etc.
Cequence Security walks an MCP server setup with no code on screen. Its May 2026 AI Gateway video runs 2:45 of captured dashboards under a voiceover: an app catalogue, endpoint selection, agent personas, guardrails and authentication settings, one screen after another. Where the product's promise is setup without custom coding, the forms are the demonstration.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
Pattern 03 · Sound · Voiceover · 49 developer-tool videos, 2024 to 2026
3. A voiceover carries 42 of 49 developer-tool videos
42 of 49 developer-tool demo, launch and explainer videos uploaded between 2 January 2024 and 3 August 2026 carry a voiceover, and speech covers at least 60 per cent of runtime in 38 of 49, a median 84 per cent. The other 7 run on music alone, and they include all four Anthropic and OpenAI videos in the set. Voiced videos run a median 1:16; music-only videos 0:46. Pinecone narrates its database architecture over diagrams. Developer tools are explained aloud; the Anthropic and OpenAI launches run on music.
Companies in this group include Microsoft, Tricentis, Acurast, Pinecone, etc.
Codex says nothing aloud. The June 2026 Codex video from OpenAI runs 0:46 on music: a chat asks Codex for an event microsite, the agent says what it checks first, and a launch-readiness tracker appears as a live site. Where the product acts on typed requests, the typed request carries the script.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
Pattern 04 · Script · Script address · 42 scripts of 15 words or more
4. 29 of 42 scripts talk to you; 13 name developers
29 of 42 developer-tool scripts of 15 words or more, uploaded between 2 January 2024 and 3 August 2026, say you, your or yourself at least twice per 100 words, a median of 2.9; we, our or us reaches that rate in 4. Developers, devs, engineers, engineering teams or builders are named in 13 of 42. Counts come from caption tracks, auto-generated for 37 of the 49 videos, so a misheard word counts as heard. HPE speaks to the OpsRamp viewer as you 5.16 times per 100 words. The engineer is spoken to as you and named as a developer in fewer than a third of scripts.
Companies in this group include Amazon Web Services, Atlantic.Net, Cube, HPE, etc.
SUSE talks about itself. Its July 2024 Choice Happens spot runs 0:41, and its caption track never says you while we or our runs at 13.16 per 100 words, listing the Linux, Kubernetes management and container security it builds. Where the buyer already runs open infrastructure, the company's own record is the pitch.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
Pattern 05 · Script · Developer terms · 42 scripts of 15 words or more
5. 27 of 42 scripts use under three developer terms per 100 words
27 of 42 developer-tool scripts uploaded between 2 January 2024 and 3 August 2026 use fewer than three developer terms per 100 words; the median script carries 2.6 per 100 words and 3 distinct terms. A developer term is any word in a fixed lexicon written before screening, from API and SDK to deploy, query and pipeline. An API term appears in 18 scripts, a code, CLI, IDE, git or language term in 5, and a programming language or git in 2; 20 of 42 reach a developer term inside the first 15 per cent of their words. Akto sells API security on a script at 0.94 developer terms per 100 words. These scripts sell developer tools in few developer words.
Companies in this group include Akto, CyberArk, Deepgram, Domino Data Lab, etc.
Microsoft packs the script with developer terms. The October 2025 Microsoft Agent Framework video runs 1:03 at 13.04 developer terms per 100 words, 13 of them distinct, naming developers, open-source agent frameworks, Semantic Kernel, AutoGen, MCP and A2A. Where the product merges two open-source projects, the names of those projects are the news.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
Pattern 06 · Runtime · Runtime · 49 developer-tool videos, 2024 to 2026
6. 33 of 49 developer-tool videos run 1:30 or less
33 of 49 developer-tool demo, launch and explainer videos uploaded between 2 January 2024 and 3 August 2026 run 1:30 or less. The median runs 1:13, the middle half spans 0:55 to 1:42, and the set runs from 0:25 to 2:45: 9 run 0:45 or less, 24 run 0:46 to 1:30, 8 run 1:31 to 2:00 and 8 run longer. OpenAI shows Codex building a site in 0:46. A developer-tool demo in this set fits in 1:30.
Companies in this group include Dremio, F5, Forethought, OpenAI, etc.
Superblocks takes 2:25. Its August 2026 Superblocks 3.0 video builds a sales CRM from one prompt, sets up a database in a private cloud, scans npm packages for supply-chain issues and deploys, one step after another under a voiceover. Where the pitch is enterprise controls around generated apps, every control gets its own screen.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
The median, 1:13, sits inside the band, and 8 videos run past 2:00.
Pattern 07 · Frame · Frame and code ground · 49 developer-tool videos, 2024 to 2026
7. 29 of 49 run light or mixed frames, not dark
29 of 49 developer-tool videos uploaded between 2 January 2024 and 3 August 2026 run light or mixed frames, 23 light and 6 mixed, against 20 dark, with a median frame brightness of 132 on a 0 to 255 scale. Frame theme is computed over nine sampled frames and counts title cards and diagrams as well as code. The code surface itself was coded by eye on frames cut at code segments: code sits on a dark ground in 5 of the 6 videos whose frames show code, and 3 of those 5 average light or mixed frames. LangChain puts its evaluator code on a light ground. A dark code surface does not make a dark video.
Companies in this group include BrowserStack, CData, GitLab, LangChain, etc.
Pinecone runs close to black. Its February 2026 How Pinecone Works video runs 2:27 of magenta and violet architecture diagrams on a dark grid, at a mean frame brightness of 15 on the 0 to 255 scale. Where the product is infrastructure nobody sees, the diagram is the screen.
Source: ADVIDS Research coded set, 49 videos from 43 companies, uploaded 2 January 2024 to 3 August 2026
Neither light nor dark holds half the set; the ground is chosen video by video.
Benchmarks
Descriptive figures for placing your own software demo video against the set. Each names its base.
Length, sound, script and screen
| Measure | Value | What it means |
|---|---|---|
| Runtime, median | 1:13 | Half of the 49 videos run shorter |
| Runtime, middle half | 0:55 to 1:42 | The span of the middle 50 per cent of the 49 |
| Speech share of runtime, median | 84 per cent | Detected speech, among videos whose audio analysis measured it |
| Words per script, median | 150 (n=42) | Caption words in scripts of 15 words or more |
| Developer terms per 100 words, median | 2.6 (n=42) | Lexicon matches per 100 caption words |
| Distinct developer terms, median | 3 (n=42) | Different lexicon terms in one script |
| You per 100 words, median | 2.9 (n=42) | You, your, you're, you've, you'll, yourself |
| We per 100 words, median | 0.0 (n=42) | We, our, us |
| Frame brightness, median | 132 | Mean luminance across nine sampled frames, 0 to 255, all 49 |
| Code share of runtime, median | 16 per cent (n=8) | Union of code segments, among visually coded videos with code on screen |
| First code on screen, median | 0:09 (n=8) | Start of the first code segment |
What fewer than half do
| Coded value | Videos | Base |
|---|---|---|
| Developer term in the first 15 per cent of words | 20 | 42 |
| Dark frames | 20 | 49 |
| Confirmed code text on screen | 5 | 13 |
| Code over a quarter of runtime | 3 | 8 |
| Kinetic type beside code | 3 | 8 |
None of these reaches half its base, so none is a pattern. A developer term inside the first 15 per cent of words, at 20 of 42, and dark frames, at 20 of 49, come closest; no visually coded video puts a presenter on camera.
What to do with this
- Pattern 01 If screen capture builds 8 of 13 visually coded developer-tool videos, prepare a demo repository and account fit to be recorded before the edit is booked.
- Pattern 02 If code is on screen in 8 of 13, with 4 from Anthropic and OpenAI, decide at the brief whether your engineers need to see the terminal or the result.
- Pattern 03 If 42 of 49 carry a voiceover, write the script for a narrator and budget the voice at the brief.
- Pattern 04 If 29 of 42 scripts speak to you and 13 name developers, decide whether your script names its engineer or addresses one.
- Pattern 05 If 27 of 42 scripts stay under three developer terms per 100 words, list the product and protocol names the script must carry and cut the rest.
- Pattern 06 If 33 of 49 run 1:30 or less, cut the demo to 1:30 and move setup steps to a separate walkthrough.
- Pattern 07 If code sits on a dark ground in 5 of 6 while 29 of 49 videos run light or mixed frames, choose the editor theme and the video's ground as two separate decisions.
Methodology
| Product category | Videos |
|---|---|
| Data platforms, databases and pipelines | 12 |
| Model, agent and LLM-application platforms | 11 |
| Build, test and deploy (DevOps, CI/CD, testing) | 7 |
| AI coding and app-building tools | 5 |
| Application, API and ML security | 5 |
| APIs, gateways and integration | 4 |
| Cloud, compute and infrastructure | 3 |
| Observability and IT operations | 2 |
| Coding scheme | Values | Definition |
|---|---|---|
| Product category | 8 categories | Judged at screening from title, description and script; the base for engineering buyers. All {n}. |
| Sound | voiceover · music only · speech share | From audio analysis. Speech share is the share of runtime with detected speech. All {n}. |
| Script address | you per 100 words · we per 100 words · names developers or engineers | Counted on caption tracks of 15 words or more. {n_scripts} scripts. |
| Developer terms | per 100 words · distinct terms · first 15 per cent of words | Every match of the fixed lexicon listed in the methodology. {n_scripts} scripts. |
| Runtime | seconds | Read from the file. All {n}. |
| Frame | light · dark · mixed · brightness 0 to 255 | Computed: mean luminance across nine sampled frames, title cards and diagrams included. Not the editor theme. All {n}. |
| Build | screen_recording · ui_mockup · 2d_animation · 3d_cgi · kinetic_type | Coded from frames, with a confidence of high or medium. {n_visual} visually coded videos. |
| Code on screen | terminal segment · confirmed code text · screen-capture segment | Coded per timed segment; code text matched by pattern and confirmed by reading; share of runtime is the union of code segments. {n_visual} videos. |
| Code ground | dark · light · both | Coded by eye on frames cut at the midpoint of each code segment. {n_ground} videos whose frames show code. |
| Not coded | cast · surface · altitude · build on the other rows | The {not_visual} videos without a visual record carry no frame coding, and this report does not code them from scripts. |
Cite this report
Jai Ghosh, Advids. 49 Developer-First Demo Videos Analyzed: How to Visually Present Code Environments for Engineering Buyers. 14 September 2026. https://advids.co/blog/developer-tool-demo-videos-code-on-screen
About Advids
Advids produces video for AI, SaaS and DeepTech companies. ADVIDS Research publishes coded studies of what those companies ship, with the dataset attached so every count can be checked.
Image credits and licensing
Video frames on this page are reproduced for analysis and commentary. Copyright in each frame remains with the company credited in its caption; to license a frame, contact that rights holder. Advids grants no rights over them.
Charts and images credited ADVIDS Research, and the coded dataset behind them, are © 2026 Advids and licensed CC BY 4.0: reuse them with attribution to Advids and a link to this page.