<- all tokdocs
2026.09.15@alberta.nyc

Open Executive does default to Claude Sonnet 5 as the video says, but its README routes strategy, finance, legal and board questions to Claude Opus 5, so the "CEO decisions are not opus level" punchline is backwards

AI
Watch on TikTok

View on TikTok ->

A 25.17 s MP4 (mov,mp4,m4a,3gp,3g2,mj2 container, 2,517,751 bytes, overall 800,238 bps) carrying an HEVC Main video stream at 1080x1920, hvc1 tag, 29.97 fps nominal with a 2,250,747/75,400 average (29.85 fps), 728,947 bps, 751 frames, 25.158492 s, alongside an HE-AACv2 audio stream at 44,100 Hz stereo, 64,472 bps, 546 frames, 25.170000 s; yt-dlp reports format id bytevc1_1080p_800206-1, SDR, aspect ratio 0.56. Posted by uploader alberta.nyc (uploader_id 6757499075077555206, channel nickname "Alberta Tech") at epoch 1789504079, which is 2026-09-15 20:27:59 UTC, with audio credited to "original sound" by Alberta Tech. The description is 35 characters and is byte-identical to the title: “AI can replace any job” okay bet 😀. At the metadata capture instant (epoch 1792408758, 2026-10-19 11:19:18 UTC) the counts stood at 116,300 views, 9,392 likes (8.08% of views), 163 comments, 1,942 reposts and 2,402 saves. I read all 13 extracted frames, numbers 001 through 013, which sit at roughly 0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22 and 24 seconds, against a transcript of 100 words in 6 segments. Frames 001 and 002 open on a single woman in clear-framed glasses with pink-tipped brown hair, seated in front of a maroon cube shelf holding a globe lamp, a green circuit-board object, a LEGO cherry-blossom bonsai and a shelf of books whose spines read "Careless People" and "AI SNAKE OIL", with a beige vintage monitor at frame left and a grey cubicle panel behind; a static two-line title card reads "How to replace your CEO with AI" and auto-captions run "about your boss" then "you can replace". Frame 003 cuts to a full-screen GitHub mock with the GitHub wordmark, a README / Code of conduct / Contributing tab row, an orange-highlighted "Open Executive" H1, badges reading "CI passing", "License Apache 2.0", "python 3.11+" and a cropped "Nex", body text beginning "An AI system that acts as your company's virtual executive" and a "Demo" heading over a card captioned "Open Executive — by Sente Lab" and "OPEN SOURCE · APACHE-2.0 · SE". Frames 004 and 005 overlay a fake macOS dialog titled "Remove your CEO…" with a businessman icon and Cancel / "Move to Bin" buttons, the cursor landing on "Move to Bin" in frame 005. Frame 006 overlays two chat bubbles, "What's the best strategy to grow my business online?" and a reply "Focus on consistent content, solve a real problem, and build trust." carrying a blue "MBA-level" verified badge. Frame 007 is clean speaker-only on the caption "qualification". Frame 008 reads "is it defaults". Frame 009 overlays a mocked settings control labelled "Model" with the value "Claude Sonnet 5" and an open dropdown listing "Claude Opus", a checked "Claude Sonnet 5" and "Claude Haiku". Frame 010 overlays a two-column card, "Opus / Advanced Reasoning" with "Complex problem solving", "Deeper analysis", "Larger context window", "Best for research & strategy", beside "Sonnet / Everyday Tasks" with "Fast & efficient", "Great at coding", "Handles most tasks", "Ideal for daily use", under a hand-drawn arrow labelled "CEO tasks" pointing at the Sonnet column. Frames 011 and 012 show the README card again at legible size: "Open Executive", badges "CI passing", "License Apache 2.0", "python 3.11+", "Next.js 15", and the full description "An AI system that acts as your company's virtual executive team — a senior advisor with Harvard MBA-level knowledge, customized for your specific business.", with the tab row now reading README / Code of conduct / Contributing / License / Security. Frame 013 closes clean on "the comments." The spoken argument is a four-beat joke: your boss may replace you with AI, so replace him first using an open source repo called Open Executive, which claims Harvard MBA level knowledge and defaults to Claude Sonnet 5, because CEO decisions do not need Opus. No terminal, no shell, no installed instance and no live product UI appears in any of the 13 frames; every interface shown is a motion-graphic mock, and the only authentic artifact on screen is the rendered README. The transcript's final segment ends at 25.04 s against an audio stream of 25.17 s and an audio.mp3 of 25.234286 s, so every word is inside the real audio and nothing in this transcript is a Whisper hallucination past the end of the file.

The repo is real and correctly named, but it was 96 days old at post time, not new

At 3.98 s the video says "There is a new open source GitHub repo called Open Executive that allows you to replace your CEO with AI." The repo exists at github.com/SenteLabsAI/OpenExecutive. The GitHub REST API reports created_at of 2026-06-11T21:41:54Z. The video posted on 2026-09-15, which is 96 days after the repository was created. "New" is doing work the timeline does not support.

The on-screen attribution is also slightly off from the organisation's own name. Frame 003 renders a demo card reading "Open Executive — by Sente Lab", singular. The GitHub organisation is SenteLabsAI and the README credits "sentelabs.ai", plural Labs. The product name itself is written two ways: "Open Executive" with a space in the README H1 and the video's title card, and OpenExecutive as one word in the repository slug.

The badges in frames 011 and 012 match the README byte for byte. The README's badge block renders CI status, "License: Apache 2.0", "Python 3.11+" and "Next.js 15", and the GitHub API confirms license.spdx_id of Apache-2.0 and language of Python. As of the 2026-10-19 capture the repo carried 5,629 stars, 624 forks, 35 watchers and 8 open issues.

"Defaults using Claude Sonnet 5" is accurate against the README as it stood on the post date

At 12.56 s the video says "my favorite part is it defaults using Claude Sonnet 5 under the hood." This checks out, and checking it requires reading the README from the right date rather than today's. The last commit touching README.md before the video posted is 5bd5f2824c4f, dated 2026-09-11T14:42:01Z. In that revision, line 60 of the configuration table reads:

| Default model | claude-sonnet-5 (Executive + most specialists) |

and line 240 of the environment-variable table reads:

| DEFAULT_MODEL| No |claude-sonnet-5 | Executive + most specialists |

The architecture diagram in that same revision labels the top-level component "Executive Orchestrator (claude-sonnet-5)". The video's claim is verified.

It is also already stale. The current main README sets DEFAULT_MODEL to claude-sonnet-5-5 and labels the orchestrator "Executive Orchestrator (claude-sonnet-5-5)". Anthropic's API release notes date the Claude Sonnet 5.5 launch to September 28, 2026, thirteen days after this video posted, and the models overview page now lists Claude Sonnet 5 under "Legacy models (still available)" alongside Claude Opus 5. Anyone who clones the repo after watching this video gets a Sonnet 5.5 default, not the Sonnet 5 default the video describes.

The frame 009 dropdown is a mock, not the product. It lists three unversioned-or-partly-versioned options, "Claude Opus", "Claude Sonnet 5" and "Claude Haiku". The README at the post date used three distinct pinned IDs: claude-sonnet-5 as the default, claude-opus-5 for deep reasoning, and claude-haiku-4-5 for intent routing and for the background episodic-memory extraction pass. No screen in this video shows the repo's actual Agent Council UI.

The Opus punchline is backwards: the README routes the strategy, finance, legal and board work to Claude Opus 5

At 17.16 s the video says "because most CEO decisions are not opus level anyway," and frame 010 reinforces it with a hand-drawn arrow labelled "CEO tasks" pointing away from the Opus column and at the Sonnet column. The repo's own configuration contradicts this. The 2026-09-11 README, line 61:

| Deep reasoning | claude-opus-5 (CSO, CFO, GC, Board — with extended thinking) |

with the matching environment variable on line 241: | DEEP_REASONING_MODEL| No |claude-opus-5 | CSO, CFO, GC, Board |.

Those four roles are the Chief Strategy Officer, the Chief Financial Officer, the General Counsel and the board-communications specialist. Competitive strategy, financial modelling, legal exposure and what you tell the board are a fair description of what a chief executive actually decides. Open Executive sends exactly those to Opus and keeps Sonnet for the orchestration layer and the remaining specialists. The video's joke inverts the repo's own routing table. The current main README preserves the same split at the newer generation: claude-sonnet-5-5 default, claude-opus-5-5 for "CSO, CFO, GC, Board — with adaptive thinking".

The video's framing of what each model is for also reads as a general description rather than one tied to the repo. Frame 010's Opus column claims a "Larger context window" than Sonnet. Anthropic's models overview page lists a 1M token context window for Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 alike, and a 128K max output for all four. On the current lineup the context window does not separate Opus from Sonnet.

"Harvard MBA level knowledge" is the repo's own marketing sentence, and what sits behind it is 96 Markdown files

At 8.26 s the video says the repo "reports having Harvard MBA level knowledge, which is probably a better qualification than your CEO has." The sourcing here is honest in one respect and worth flagging in another. The video says "reports," and the repo does report it: the 2026-09-11 README line 8 is verbatim the sentence burned into frames 011 and 012, "An AI system that acts as your company's virtual executive team — a senior advisor with Harvard MBA-level knowledge, customized for your specific business." The video is accurately quoting a README.

That sentence is a self-description with no measurement behind it, and nothing in the repository ties it to Harvard. The README's technical section is more modest, calling the corpus "built-in MBA-level Markdown (knowledge/builtin/, git-tracked) seeded into ChromaDB at startup." I pulled the full file tree at commit 5bd5f2824c4f: packages/core/openexecutive/knowledge/builtin/ holds 96 Markdown blobs totalling 666,740 bytes, split across ten folders, strategy (9), finance (9), product (9), hr (8), legal (8), operations (8), marketing (7), board (6), skills (15) and a failures corpus of 17 case studies including enron.md, theranos.md, ftx-collapse.md, boeing-737-max.md, wework-ipo.md, quibi.md and kodak-digital.md. That is a hand-written reference library of business case material, which is a real and specific thing, and it is not a Harvard credential. I found no Harvard affiliation, licence or curriculum attribution anywhere in the repository tree. The "Harvard MBA-level" descriptor, and the blue "MBA-level" verified badge the video stamps on a chat reply in frame 006, are rhetorical. I mark both Unverified.

The repo does carry one measured quality bar, and the README states it plainly: "evals/ contains 29 scenarios covering all 8 domains, scored by claude-opus-4-7 as an LLM-as-judge... Five scoring dimensions (persona coherence, domain accuracy, company context utilization, routing quality, actionability) are each rated 1–5. The CI gate requires ≥ 3.5/5 average." Twenty-nine scenarios graded by another Claude model to a 3.5-out-of-5 floor is a self-consistency gate. It is not evidence about executive decision quality, and the video does not mention it.

The README is also internally inconsistent about how many specialists there are. Line 23 says "a single coherent executive voice backed by nine specialist AI agents." The GitHub repo description says "8 specialist agents." The directory comment at line 86 says "# 9 specialist agents." The MCP section at line 305 says "any of the ten specialists." Four counts, one repo.

Measured evidence does not support "AI can replace any job," including for the job the video targets

The title's claim, “AI can replace any job” okay bet, is the premise the whole joke rides on, and three dated measurements push back on it.

The METR randomized controlled trial is the cleanest. "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity" by Joel Becker, Nate Rush, Elizabeth Barnes and David Rein was submitted to arXiv on July 12, 2025 as 2507.09089, revised July 25, 2025. Sixteen developers with moderate AI experience completed 246 tasks in mature repositories where they averaged five years of prior experience, with each task randomly assigned to allow or disallow AI tooling. The abstract states the result: "Before starting tasks, developers forecast that allowing AI will reduce completion time by 24%. After completing the study, developers estimate that allowing AI reduced completion time by 20%. Surprisingly, we find that allowing AI actually increases completion time by 19%." Economics experts had predicted 39% shorter and ML experts 38% shorter. On the one job where AI assistance is most mature and most measured, a controlled trial found a 19% slowdown, and the practitioners themselves could not tell, misjudging the sign by roughly 39 percentage points.

The Bureau of Labor Statistics does not project the executive job disappearing. The Occupational Outlook Handbook page for Top Executives, last modified August 27, 2026, gives 3,890,600 jobs in 2025, median pay of $108,780 per year or $52.30 per hour, and a 2025 to 2035 outlook of "5% (Faster than average)" with an employment change of 190,600. The page states: "Overall employment of top executives is projected to grow 5 percent from 2025 to 2035, faster than the average for all occupations. About 304,100 openings for top executives are projected each year, on average, over the decade." The federal projection for the specific occupation this video proposes deleting is growth above the all-occupation average.

Anthropic's own usage data points the same way. The Economic Index report "Cadences," published June 26, 2026, drawing on roughly 9,700 survey respondents with linked usage data over April 10 to June 10, 2026, reports that "Management, at 23% of respondents, is also heavily over-represented relative to its 7% employment share, even though it accounts for only 4% of sessions," and adds the interpretation: "This gap is consistent with managers using Claude for tasks other than management itself." Managers are among the heaviest users of Claude and among the lightest users of it for managing. That is the opposite of the substitution the video stages.

The two-speaker hypothesis fails: one person, one continuous take, no mirrored second character

I was asked to check whether the two speakers in this video are the same person flipped, on the pattern of this creator's other video in the corpus. There are no two speakers. Across all 13 frames there is one woman, one camera position, one framing, one wardrobe (black sleeveless top), and one lighting setup, with the maroon shelf, globe lamp, circuit-board object and LEGO bonsai in consistent left-to-right order in every shot. The books stay in the same order with "AI SNAKE OIL" at the right in frames 001, 002, 004, 005, 006, 008, 009, 010, 011, 012 and 013. Nothing is mirrored and nothing is flipped. The only cut away from the speaker is frame 003, the full-screen GitHub mock at roughly 4 s. This video uses a single-presenter direct-address format with graphic overlays, not a two-character format.

The rebuttal framing in the brief also does not match the artifact. The title sets up "AI can replace any job" as a claim to be answered, but the spoken content never rebuts it. The video accepts the premise and extends it upward to the CEO as a joke. The closing line at 21.94 s, "if you successfully replace your boss with this, let me know in the comments," is a comment-bait call to action, not an argument.

Key Takeaways

  • Verified: The repo exists at github.com/SenteLabsAI/OpenExecutive, is Apache-2.0 licensed, is Python-first with a Next.js 15 frontend, and at the 2026-10-19 capture had 5,629 stars and 624 forks.
  • Verified: At 12.56 s the video says the repo "defaults using Claude Sonnet 5." The README at commit 5bd5f2824c4f, dated 2026-09-11 and the last README commit before the post, sets DEFAULT_MODEL to claude-sonnet-5.
  • Correction: At 17.16 s the video says "most CEO decisions are not opus level anyway," and frame 010 points a "CEO tasks" arrow at the Sonnet column. The same README sets DEEP_REASONING_MODEL to claude-opus-5 for "CSO, CFO, GC, Board," which is strategy, finance, legal and board communications. The repo routes the most CEO-like decisions to Opus.
  • Correction: At 3.98 s the video calls it "a new open source GitHub repo." The GitHub API reports created_at of 2026-06-11T21:41:54Z, making the repo 96 days old on the post date.
  • Partial correction: Frame 010 claims Opus has a "Larger context window" than Sonnet. Anthropic's models overview lists 1M tokens of context and 128K max output for Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 alike.
  • Unverified: "Harvard MBA level knowledge," quoted at 8.26 s and burned into frames 011 and 012, is the README's own marketing sentence. I found no Harvard affiliation anywhere in the repository tree. What backs it is 96 Markdown files totalling 666,740 bytes across ten domain folders.
  • Unverified: The blue "MBA-level" verified badge stamped on the chat reply in frame 006 is a motion-graphic flourish with no measurement behind it. No terminal, shell or live product UI appears in any of the 13 frames.
  • Clarification: The default has already moved. Current main sets DEFAULT_MODEL to claude-sonnet-5-5, and Anthropic's release notes date Claude Sonnet 5.5 to September 28, 2026, thirteen days after this video posted.
  • Clarification: The two-character mirrored-self hypothesis is wrong for this video. All 13 frames show one speaker, one camera position and an unmirrored background, with a single cut to a full-screen GitHub mock at roughly 4 s.
  • Context: METR's RCT (arXiv:2507.09089, submitted July 12, 2025) found that allowing AI tools increased task completion time by 19% across 16 developers and 246 tasks, while those developers forecast a 24% reduction and estimated a 20% reduction afterwards.
  • Context: The BLS Occupational Outlook Handbook, last modified August 27, 2026, projects top executive employment growing 5% from 2025 to 2035, a change of 190,600 jobs from a 2025 base of 3,890,600, with about 304,100 openings per year.
  • Context: Anthropic's Economic Index "Cadences" report, published June 26, 2026, found management at 23% of surveyed respondents against a 7% employment share but only 4% of sessions, which it reads as "managers using Claude for tasks other than management itself."
  • Context: The repo's only quantitative quality bar is 29 eval scenarios across 8 domains, judged by a Claude model against a 3.5-out-of-5 CI floor on five dimensions. That measures self-consistency, not executive decision quality, and the video does not mention it.

Resources

Published September 15, 2026. Writeup generated from a favorited TikTok.