How I AIMastering ChatGPT: Advanced techniques for workplace communication and productivity | Hiten Shah
CHAPTERS
- 0:00 – 5:23
Practical tips for using ChatGPT well: memory hygiene, temp chats, and limits
Claire asks for Hiten’s best practices beyond “be thoughtful about memories.” Hiten explains how he separates disposable work into temporary sessions, archives chats to reduce memory bleed, and why higher-tier ChatGPT access changed his day-to-day usage compared with Claude’s limits.
- •Use temporary chats for anything you don’t want remembered
- •Archive/unarchive chats to manage what influences future interactions
- •Why the $200/month plan shifted usage (fewer limits, more Projects)
- •Claude vs ChatGPT tradeoffs: rate limits and reliability when you need results fast
- 5:23 – 8:00
Context is everything: ask for needed context and show “what great looks like”
Hiten lays out his core philosophy: don’t expect magic from a one-line prompt without context. He emphasizes explicitly asking the model what context it needs and, most importantly, providing examples of excellent output to train the model toward your desired standard.
- •Front-load context or plan to build it over time
- •Ask the model: “What context do you need?”
- •Use frameworks the model already understands to guide outputs
- •Feed exemplar outputs (AI or human) to “codify” quality and scale it
- 8:00 – 10:18
Demo setup: creating a ‘What Would Morgan Do?’ boss-simulation Project
They build a project to simulate Hiten’s boss (Morgan) using Morgan’s operating manual and an article Morgan likes. Hiten shows how he creates project instructions so ChatGPT can respond in Morgan’s style and offer manager-appropriate feedback.
- •Create a dedicated Project for a specific person/use case
- •Upload key artifacts (operating manual, favorite writing) as files
- •Write instructions to simulate feedback/advice in that person’s voice
- •Goal: prep for real conversations by rehearsing how your boss will react
- 10:18 – 12:34
Iterating on instructions: ‘AI shaping’ and turning good outputs into reusable templates
Hiten demonstrates how he pushes ChatGPT to produce paste-ready project instructions, then refines when the first attempt is incomplete. He explains a repeatable pattern: when an output is good, immediately have the model codify it so you can reuse the structure later (even making a ‘project for projects’).
- •Prompt the model to generate instructions specifically formatted for Projects
- •When output is off, ask for the exact artifact you need (paste-ready instructions)
- •Maintain a meta-doc/process for ‘shaping’ model behavior
- •Codify good outputs into reusable templates to avoid redoing work
- 12:34 – 14:16
Using the boss Project: crafting a pitch your manager will actually buy
With the boss-simulation project in place, Hiten asks how to pitch a wild product idea to Morgan. The output reflects Morgan’s preferences (how to structure the pitch, what he respects, and the questions he’ll ask), illustrating how Projects enable more tailored communication guidance.
- •Prompt: best way to pitch a bold idea to a specific manager
- •Output becomes actionable: structure, framing, and anticipated reactions
- •Value comes from combining instructions + uploaded artifacts
- •Takeaway: ICs can ‘replicate their boss’ to manage up effectively
- 14:16 – 16:46
Personality frameworks as coaching fuel: Enneagram, Myers-Briggs, Human Design
The conversation shifts to using personality frameworks to improve communication and relationships at work. Hiten explains how adding types (e.g., Morgan as Enneagram 5, Hiten as 9) can help the model suggest better ways to meet in the middle—then Claire shares her own team’s practice of collecting personality/work-style data.
- •Personality tests can make guidance more relationship-specific
- •Enneagram emphasized for relationship/communication dynamics
- •Claire’s team collects DISC/Enneagram/Myers-Briggs + a ‘10 questions about me’ doc
- •Idea: store team knowledge in a repository and query it for conflict resolution coaching
- 16:46 – 20:26
Building a ‘personal OS’ Project to coach yourself in real situations
Hiten introduces his ‘personal operating system’ project—an always-on coaching environment built from instructions and personal artifacts (like a voice/tone guide and personality data). He shows how a simple prompt produces structured advice, even generating practical templates like a weekly check-in format aligned to a manager’s style.
- •Personal OS = self-coaching + operating guidance in a dedicated Project
- •Use instructions that define purpose, behaviors, and suggested files
- •Upload personal artifacts (tone guide, assessments) to personalize responses
- •Outputs can include actionable formats (1:1 updates, check-ins, unblocker framing)
- 20:26 – 20:57
Why Projects beat plain chats: grounded quotes, stronger recall, better specificity
Hiten highlights that Projects produce better, more referential output because they can quote and anchor on uploaded documents. He contrasts this with generic chat behavior where the model tends to “fill in blanks,” and he notes he prefers Projects over custom GPTs for this kind of grounded workplace use.
- •Projects can cite/quote the source manual directly
- •Improves specificity vs generic advice in normal chats
- •Easier to maintain evolving context than one long conversation
- •Custom GPTs are optional; Hiten prefers Projects for this workflow
- 20:57 – 23:45
Mixing frameworks + personal context: creating affordable ‘always-on’ coaching
Claire reflects on the broader pattern: structured frameworks (personality models, workplace dynamics) plus individual context enables consistent coaching for common scenarios (hard conversations, conflict, pitching). They compare this to expensive executive coaching (360 feedback + frameworks) and show how ChatGPT can approximate parts of that process.
- •Frameworks provide structure; personal context makes it relevant
- •Common workplace scenarios become ‘queryable’ coaching moments
- •Parallel to coaching workflows: 360s, pattern interruption, reframing
- •Example: emotions + workplace politics prompt yields insight (control vs integrity)
- 23:45 – 27:55
Sales enablement Project: implementing Winning by Design with PDFs
Hiten demonstrates a second major use case: loading the Winning by Design sales framework into a Project using publicly available PDFs. He shows how to ask the Project what it can do, then generate a SPICE discovery guide—turning static enablement docs into a dynamic, reusable assistant.
- •Gather framework assets (PDFs) and load them into a Project
- •Ask the Project: “What can you help me with?” to discover capabilities
- •Generate a SPICE discovery guide aligned to the framework
- •Value: framework-based questioning surfaces non-obvious discovery prompts
- 27:55 – 31:38
Improving outputs with deep research + memory pruning + fast re-prompts
They refine the sales output by adding product context via a deep research prompt and re-running the task. When the model misinterprets instructions, they embrace the speed of iteration: re-prompting is cheap, mistakes can be instructive, and memory updates can be pruned when they’re not helpful.
- •Use deep research to enrich context (company/product specifics)
- •Copy/paste research into the Project to improve deliverables
- •When it’s wrong, rerun quickly—iteration beats perfectionism
- •Watch memory updates and prune/remove what you don’t want retained
- 31:38 – 36:18
Deep research on Claire: building a ‘boss profile’ and testing advice quality
Hiten runs deep research on Claire, then creates a new Project with that research plus Claire’s personality details (Enneagram, Myers-Briggs, DISC, astrology). They test a practical scenario—how to tell Claire “no” on a roadmap item—and Claire evaluates what rings true versus what misses (like her dislike of rigid frameworks).
- •Create a Project from internet research + explicit self-reported traits
- •Test a realistic prompt: saying ‘no’ to a leadership request
- •Validate what advice matches real preferences (directness, experiments)
- •Note limitations: research can be ‘dirty’; some guidance remains generic
- 36:18 – 40:11
Lightning round: tool stack, ‘blank canvas’ advantage, and don’t race to automation
Hiten shares he uses AI 3–6 hours daily, mostly ChatGPT and Claude, plus a private desktop tool. His core product/ops belief: people rush into automation before they understand prompts; instead, use manual chat work (‘bare metal’) to iterate until outputs are reliable, then automate.
- •Daily AI usage habits and primary tools
- •Why ChatGPT’s blank canvas still wins for experimentation and iteration
- •Manual repetition builds the plan for automation; automation too early locks in bad outputs
- •Model choice and reliability matter; treat prompt design like eval-driven iteration
- 40:11 – 42:53
How to correct ChatGPT: be blunt, avoid bribery, and think incentives
To close, Hiten explains his debugging style: directly state what’s incorrect, propose an assumption about what went wrong (often instructions), and request a fix. He discourages bribing tactics, arguing that if you treat the model like a human, incentives matter—especially with memory-enabled behavior shaping.
- •Prompting technique: “This is incorrect—here’s what I think is wrong”
- •Focus on fixing instructions and constraints rather than gimmicks
- •Avoid bribery tricks; they create perverse incentive patterns
- •Final takeaway: if you anthropomorphize, do it thoughtfully around incentives
