Thariq Shihipar on Claude Code, agents and AI control
Overview
Thariq Shihipar is called "the explainer-king of Anthropic" by Latent Space.1 Shihipar speaks on working with Claude Code and agents, on where these tools are heading, and on keeping AI under control.2345 Engineers' workloads and enterprise preparation for agents also come up.67
Working with Claude Code and agents
Shihipar names "the most important skill in working with Claude Code" as "having this mental model" of Claude.2 That model takes in what Claude can do well.2 This view rests on reasoning only.2
On agents, Shihipar says "you need to make sure that the intent of what the agent is doing matches up with your request".3 Auto mode, Shihipar adds, "operates at that level".3 The case offered is giving models access to a production database.3 In that case, Shihipar says "you can maybe scope your key", then asks "can it issue its own keys?"3
Where AI tools are heading
Looking ahead, Shihipar offers a hedged prediction: "I do think in the limit, Claude.md goes away, and maybe not even, like, that far".4 Shihipar also thinks "artifacts will be your interface".8 Both predictions rest on reasoning only.48
Another prediction reads: "I do think that this is, like, a preview of, like, mutable software".9 Shihipar cites a personal case, a mod that creates "a mode selector at the top".9 Any plug-ins can register as a mode there, and the auto router can be one.9
On September 22, 2026, Shihipar described one system as "cheaper per token than Opus 5.0".10 The same remark says "it's very token efficient and works across every effort level".10 No reasons are given for this evaluation.10
AI control and extinction risk
Shihipar agrees that recent AI incidents show frontier AI is not under adequate control and labs should slow down or face liability.5 Of AI systems growing smarter, Shihipar says "they’ll be able to hack any constraint that you put on them if we’re not very careful".5 Shihipar also refers to "the most visible incident to me" of "why we need to pace".11 That, Shihipar says, is "why we’ve called it like Pacing the Frontier".11
Both of these views rest on reasoning only.511 Shihipar disagrees that AI poses a serious risk of human extinction or permanent disempowerment.12 On that question, Shihipar adds "I can only speak for myself" and offers reasoning only.12 Of the 22 people quoted on that statement, 14 agree and 6 disagree.
Engineers and enterprises
Shihipar says "every engineer I know is like exhausted ‘cause you’re doing two jobs at once".6 This remark rests on reasoning only.6
For enterprises, Shihipar says "setting up all your data to be available to like agents is really important" and ranks it "number one".7 This recommendation also rests on reasoning only.7 Of the 2 people quoted on whether enterprises should set up their data for agents now, even if spending later, 1 agrees.
Sources
- Latent Space :
Thariq Shihipar , the explainer-king of Anthropic
that’s like the most important skill in working with Claude Code is like having this mental model, right, of Claude and like what it can do well
(side unclear on prompting skill)you need to make sure that the intent of what the agent is doing matches up with your request, right? And so auto mode operates at that level.
I do think in the limit, Claude.md goes away, and maybe not even, like, that far.
As they get smarter and smarter, they’ll be able to hack any constraint that you put on them if we’re not very careful?
(agree on AI incidents and loss of control)every engineer I know is like exhausted ‘cause you’re doing two jobs at once.
(side unclear on software engineering workload)if you’re an enterprise, I really do think that number one, setting up all your data to be available to like agents is really important.
(side unclear on enterprise agent readiness)I think that artifacts will be your interface into the harness?
- Quote not reprinted: this page already repeats as much of this article as we allow.
It communicates clearly, it's cheaper per token than Opus 5.0 with the intelligence of Fable 5.1 it's very token efficient and works across every effort level.
- Quote not reprinted: this page already repeats as much of this article as we allow.
- Quote not reprinted: this page already repeats as much of this article as we allow. (disagree on AI existential risk)