Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
The AI industry just invented a new job. Wait until you hear what it does.
TL;DR
Anthropic’s “forward deployed engineer” role is being framed as the new AI-age résumé hack — Mo Bitar says the clearest signal of what companies want right now is Anthropic’s own posting for “Forward Deployed Engineer, Applied AI,” and he bluntly tells viewers to mirror its responsibilities on their résumés.
He treats the job title as a confession that AI still needs humans embedded on-site to make it work — instead of the promised autonomous software revolution, Bitar argues companies are buying elite implementation people who build flashy Claude-powered demos, MCP servers, and sub-agents inside customer systems.
Bitar’s core critique is that labs are selling “humans as a service,” not labor-replacing AI — he points to Anthropic’s reported $1.5 billion joint venture with Blackstone and Goldman Sachs and OpenAI’s alleged $14 billion “DeployCo” as evidence of a services-heavy model.
His dystopian scenario is software development consolidating into a few model providers — in his telling, companies stop hiring engineers directly and instead rent forward-deployed talent from Anthropic or OpenAI, while the best software-building models stay tightly gated behind products like “Mythos.”
His practical advice is aggressively tactical: optimize for the AI-screened hiring funnel — copy the language from the job post, study AI concepts like agent orchestration the way candidates once grinded LeetCode, and use Claude live in technical interviews to produce impressive demos fast.
He argues interview success in the AI era is as much performance as engineering — invoking Dwarkesh’s prep habits and the story of Soham working 12 jobs, Bitar says the winners will be the people who can sound fluent, ship a convincing prototype, and let the model do most of the work.
The Breakdown
The New Buzzword to Put on Your LinkedIn
Bitar opens with a joke that lands like advice: forget “prompt engineer” or whatever last month’s AI title was, and change your LinkedIn headline to “forward deployed engineer.” He says nobody really knows what it means yet, which is exactly why it works — it sounds elite, provocative, and very now.
Anthropic, Palantir, and the Rise of “Humans as a Service”\n
He explains that “forward deployed engineer” comes from Palantir, where engineers get embedded inside customer organizations. In Bitar’s retelling, Anthropic and OpenAI are now effectively doing the same thing at huge scale — he cites Anthropic’s $1.5 billion joint venture with Blackstone and Goldman Sachs and OpenAI’s $14 billion “DeployCo” — which leads to his punchline that the AI labs have somehow become temp agencies for humans.
The Navy SEAL Fantasy and the Slop Reality
Bitar leans hard into the bit: these engineers are the “Navy SEALs of AI,” black T-shirts, helicopters, instant action. But the actual work, he says, is building demos and prototypes that look incredible in week one, wow the execs, and only reveal themselves as “slop” months later when real usage hits and the original team has already disappeared.
His Bigger Claim: AI Didn’t Replace Labor, It Created Dependence
This is the heart of the rant. He says AI companies promised autonomous systems that companies could simply plug in, but what they’ve actually delivered is a tool that makes internal teams dependent on outside experts — then sold those experts back as the cure.
The Dystopian Endgame: Anthropic as the Gatekeeper of Software
From there he spins out a bigger industry scenario: in 10 years, maybe companies don’t hire software engineers at all, they just call Anthropic or OpenAI when they need software. He argues the economics could even make sense if those internal teams get access to tokens at cost, and says the real lock-in comes when top-tier software-building models are withheld from the public while weaker, safer, or less deployable ones are exposed.
The Escape Hatch: Open Models or Permanent Dependence
Bitar says the less scary outcome depends on open-source models keeping up. If labs keep holding back their best internal coding and security-capable systems — he names “Mythos” as an example — then outsiders may only be able to build “cute little software” in-house while the serious stuff stays under vendor control.
His Very Practical Resume and Interview Playbook
Then the video shifts from industry rant to job-market hustle. He tells viewers to pull up Anthropic’s job posting, lift the responsibilities section — building production apps with Claude models, delivering MCP servers and sub-agents, staying current on LLM capabilities, managing customer relationships — and rewrite it into their own experience because that’s exactly what applicant-screening systems will reward.
Become a “Mini Dwarkesh” and Let Claude Do the Demo
For interviews, his model is pure preparation theater: study agentic systems, orchestration patterns, and abstract AI concepts the way people once studied algorithms for LeetCode. And in the technical round, assuming you can already code, he basically says to open Claude, dump everything into it, let it generate the 100-agent orchestration system on the spot, and accept the applause — because in his telling, that performance itself is now what counts as AI engineering.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
