Last week we covered Claude Code and Claude Cowork in full.
This week we do the same for OpenAI’s two most powerful tools for builders. ChatGPT, now an agentic reasoning system far beyond a chatbot, and Codex, OpenAI’s autonomous software engineering agent.
These are not tools you use occasionally for a quick answer. Used properly, they function as a reasoning partner, a research engine, a writing collaborator, a code reviewer, and an autonomous software engineer, all in one place.
The same principle from the Claude article applies here. Understanding the full surface area of what is available determines how much value you extract. Most people use ten percent of what these tools can do.
This article covers all of it. Every major feature, who it is for, and how different roles and organisations put it to work.
Part One: ChatGPT — Core Features
What ChatGPT actually is today
ChatGPT started as a conversational AI. It is now an agentic AI system.
The distinction matters. A conversational AI responds to your messages one at a time. An agentic system plans, acts across tools, reasons through multi-step problems, and executes workflows on your behalf.
ChatGPT today can search the web, read your files, execute code, connect to external services, remember your context across sessions, and run multi-step tasks from start to finish. The chatbot framing undersells what it has become.
Feature 1: Conversational reasoning
The foundation of everything ChatGPT does is its ability to reason through complex problems in conversation.
Product strategy. Technical trade-offs. Architecture decisions. API design. Customer journeys. Prioritisation frameworks. Debugging flows. Systems thinking. These are not things ChatGPT retrieves from a database. It reasons through them based on the context you provide.
The quality of that reasoning is directly proportional to the quality of your context. This is why the fundamentals articles earlier in this series matter. Understanding what you are asking for changes the quality of what you get back.
How PMs use this. Think through a product decision in real time. Give ChatGPT the problem, the constraints, the user context, and the options you are considering. Use it to stress-test your thinking before you present it to stakeholders.
How architects use this. Reason through system design trade-offs. Monolith versus microservices. REST versus GraphQL. Kafka versus RabbitMQ. Give it the context of your system and ask it to map the implications of each choice.
How founders use this. Work through strategic decisions by giving ChatGPT the full context of your situation and asking it to surface the trade-offs, the assumptions you might be making, and the questions you have not asked yet.
Feature 2: Deep research and web search
ChatGPT can search the web, synthesise recent technical developments, compare frameworks and tools, analyse competitors, and gather API documentation insights in real time.
This is not a simple web search. It searches, reads, synthesises, and returns a structured answer with sources. The difference between searching the web yourself and asking ChatGPT to search for you is the synthesis layer. You get insight, not a list of links.
How PMs use this. Market intelligence on a competitor’s recent product moves. Analysis of a new regulatory requirement. Synthesis of user research from industry reports. What previously took half a day of reading takes twenty minutes.
How architects use this. Evaluate a new framework or tool against your current stack. Get a structured comparison of approaches with real-world evidence rather than theoretical arguments.
How enterprises use this. Keep leadership briefed on industry movements without dedicating analyst time to weekly research. Ask ChatGPT to monitor and summarise developments in your sector on a regular cadence.
Feature 3: File understanding
ChatGPT can read and analyse PDFs, slides, spreadsheets, codebases, architecture diagrams, logs, requirements documents, and CSVs. You upload the file and ask your question.
The practical implication is that the documents which previously required dedicated time to read and summarise can now be processed in seconds. A 60-page requirements document. A quarterly financial report. An incident log from last month. A competitor’s published API documentation.
How PMs use this. Drop last quarter’s user research into ChatGPT and ask for the three themes most relevant to your current roadmap priority. The synthesis that normally takes a sprint retrospective happens instantly.
How engineers use this. Upload an incident log and ask ChatGPT to identify the pattern across failures. Give it a legacy codebase file and ask it to explain the architecture before you start a migration.
How enterprises use this. Process inbound documents at scale. Contract reviews, technical audits, architecture assessments, migration planning. The first layer of document analysis no longer requires a dedicated human reviewer.
Feature 4: Multimodal understanding
ChatGPT understands images as well as text. Screenshots, UI mockups, architecture diagrams, whiteboards, product flows, and code screenshots can all be uploaded and discussed.
This changes how you can use it for design reviews, system design discussions, debugging UI issues, and developer onboarding. You stop describing what you are looking at and start showing it.
How designers use this. Upload a mockup and ask for a critique of the user flow. Upload a competitor’s screenshot and ask for a feature comparison. Visual context produces much more relevant analysis than a text description of a visual thing.
How PMs use this. Photograph a whiteboard from a discovery session and ask ChatGPT to structure the output. Upload a diagram from a stakeholder meeting and ask it to identify gaps in the proposed flow.
How engineers use this. Upload an architecture diagram and ask ChatGPT to identify single points of failure or bottlenecks. Upload a UI screenshot showing a bug and ask it to reason about what might be causing the behaviour.
Feature 5: Writing and structured document generation
ChatGPT generates polished, structured documents from context. PRDs, RFCs, user stories, stakeholder updates, sprint summaries, release notes, architecture documentation, executive memos.
The quality of the output depends entirely on the quality of the context you provide. A vague brief produces a generic document. A detailed brief with specific context, constraints, and requirements produces something you can actually use.
How PMs use this. Give ChatGPT the problem statement, the user, the proposed solution, the constraints, and the success metrics. Ask it to draft the PRD. Review and refine rather than writing from a blank page.
How technical PMs use this. Generate RFCs that include architecture considerations, API design implications, and technical constraints alongside the product requirements. The technical and product layers in one document.
How enterprises use this. Standardise the quality of written outputs across teams. Every PRD, every release note, every stakeholder update follows the same structure and quality floor. The strongest writers lift the whole organisation.
Part Two: ChatGPT — Advanced Agentic Features
Feature 6: Agentic task execution
Modern ChatGPT functions as an AI agent system. It does not just respond to questions. It performs multi-step workflows, reasons across tools, executes coding tasks, browses documentation, runs calculations, analyses files, and iterates on solutions.
You give it a goal. It plans the steps. It executes them. It reports back.
This is foundational for AI-native product development, autonomous workflows, and technical operations. The shift from chatbot to agent is the most significant change in how these tools work and the one most people have not fully absorbed yet.
How PMs use this. Ask ChatGPT to research a market opportunity, synthesise the findings, draft an executive summary, and identify the three biggest open questions. One instruction. Four steps. One output.
How founders use this. Delegate an entire research and analysis workflow. Competitive landscape, user pain points, market sizing, strategic implications. The kind of work that previously occupied an analyst for a week happens in an afternoon.
Feature 7: Tool use and integrations
ChatGPT integrates with GitHub, Google Drive, Gmail, Calendar, APIs, local files, and coding environments. These connections let it act on your behalf across the tools you already use.
When ChatGPT is connected to GitHub, it can read your repositories, understand your codebase structure, and reference real code in its responses. When connected to Google Drive, it can pull documents into context. When connected to your calendar, it can schedule around your actual availability.
How engineers use this. Connect your GitHub repository and ask ChatGPT to review the last ten pull requests for recurring patterns in review feedback. Ask it to find all the places in the codebase where a particular pattern is used.
How PMs use this. Connect Google Drive and ask ChatGPT to synthesise the last three quarters of customer feedback documents into a prioritised list of product improvement opportunities.
How enterprises use this. Build a connected workspace where ChatGPT can pull from the actual sources of truth: documentation, repositories, spreadsheets, and communications, rather than relying on what someone remembers to paste in.
Feature 8: Persistent context and memory
ChatGPT can remember your projects, frameworks, goals, preferences, architecture patterns, and company strategy across sessions.
This changes the quality of every interaction. Instead of re-briefing the tool at the start of every conversation, it remembers. Your preferred output format. The constraints of your product. The decisions you have already made. The principles you work by.
How PMs use this. Set up memory for your product context. The user persona. The current strategic priorities. The technical constraints. Every PRD, user story, and stakeholder update that follows is grounded in that context automatically.
How architects use this. Maintain an evolving architecture context. Every system design discussion builds on what has already been decided rather than starting from scratch.
How startups use this. Memory becomes a kind of operating system for the company. Goals, constraints, strategy, and working principles are always in context. The AI becomes a consistent collaborator rather than a tool you brief anew each time.
Feature 9: Automation and scheduling
ChatGPT can schedule reminders, monitor tasks, run recurring workflows, and create operational automations.
The practical application is that the recurring operational work, the weekly report compilation, the sprint ritual, the stakeholder update, the governance review, can be automated rather than repeated manually.
How PMs use this. Set up a recurring synthesis that pulls last week’s user feedback, summarises the themes, and delivers a structured briefing before the weekly planning session.
How enterprises use this. Build operational cadences that run automatically. Weekly reporting. Monthly governance reviews. Quarterly competitive analysis. The rhythm of the business operates independently of someone remembering to do it.
Part Three: Codex — OpenAI’s Software Engineering Agent
What Codex actually is
Codex is OpenAI’s autonomous software engineering system, integrated into ChatGPT.
Unlike traditional autocomplete tools that suggest the next line, Codex works end to end. You describe a goal. Codex reads your repository, plans the implementation, writes the code, runs the tests, and opens a pull request ready for review.
It does not complete code as you type. It takes on tasks. The distinction is the same one we drew for Claude Code in the last article. Responding versus acting.
Feature 10: Autonomous coding
Codex can write features, fix bugs, refactor code, generate tests, migrate systems, and create PR-ready changes without you writing a single line.
Each task runs in an isolated cloud environment with repository awareness and the full dependency stack available. Codex can install packages, run scripts, and validate changes before they reach you.
How engineers use this. Delegate the implementation of a well-specified feature. Review the pull request it produces rather than writing the code from scratch. The engineering review process remains. The implementation process is accelerated.
How PMs use this. Commission working prototypes from a brief. Not mockups. Working software. Give Codex the spec and review what it produces before committing engineering resources to a full build.
How enterprises use this. Accelerate backlog execution on well-defined tickets. The engineering capacity constraint shifts from implementation to architecture and review. Teams ship faster because the implementation work runs in parallel with planning.
Feature 11: Multi-agent parallel workflows
Codex supports multiple AI agents working simultaneously on different tasks in isolated environments. One agent fixes a bug. Another writes the tests. A third updates the documentation. All in parallel.
This is the engineering capacity multiplier that is reshaping how teams think about throughput. A single engineer orchestrating multiple Codex agents across a large repository can achieve a throughput that previously required a team.
How platform teams use this. Run parallel workstreams across a large repository without context-switching between them. Each agent operates in isolation and reports back when complete.
How enterprise engineering uses this. Accelerate large, parallelisable projects. A migration, a refactor, a testing coverage expansion. Work that would have taken months of coordinated engineering effort runs in a fraction of the time.
Feature 12: Sandbox execution
Every Codex task runs in an isolated cloud environment with full repository awareness, all dependencies installed, and the ability to run the full test suite.
Codex does not just generate code for you to paste in. It validates the code in an environment that mirrors your actual setup. Tests pass before you see the pull request. Dependencies are resolved. The build does not break.
How engineers use this. Receive pull requests that arrive with passing tests and a working build rather than code that looks right but has not been run. The review is about the logic and the design, not the mechanics.
How enterprises use this. Reduce the cost of code review by ensuring a minimum quality bar is met before human review begins. The engineering governance layer focuses on architectural quality rather than catching execution errors.
Feature 13: GitHub integration
Codex integrates natively with GitHub repositories, pull request workflows, issues, branch management, and CI/CD pipelines.
You can give Codex a GitHub issue and it will fix it. You can ask it to review open pull requests and summarise the changes. You can connect it to your CI pipeline and have it respond to failures automatically.
How engineering managers use this. Monitor open pull requests, identify review bottlenecks, and accelerate merge velocity without adding reviewer headcount.
How DevOps engineers use this. Connect Codex to your CI/CD pipeline and have it respond to build failures. Read the failure, identify the cause, propose a fix, and open a pull request. The on-call burden for pipeline failures shrinks significantly.
Feature 14: Deep codebase understanding
Codex understands architecture relationships, dependencies, modules, API contracts, and system structure across large codebases. It traces how changes in one part of the system affect other parts.
This is what separates Codex from a code generation tool. A code generation tool produces code that might work in isolation. Codex produces code that works within the specific context of your system.
How engineers use this. Onboard to a new repository in minutes rather than weeks. Ask Codex to explain the architecture, trace a data flow, and identify the right place to make a change before you touch a line of code.
How enterprises use this. Make legacy system knowledge accessible. The architecture understanding that previously lived only in the heads of long-tenured engineers becomes accessible to the whole team through Codex.
Feature 15: PR and review generation
Codex opens pull requests, generates diffs, summarises changes, and explains implementation details. Every PR arrives with a clear description of what changed, why, and how the implementation works.
The documentation quality of engineering work rises as a side effect. Every change is explained because Codex generates the explanation as part of the task. Institutional knowledge gets captured automatically.
How engineering managers use this. Receive PRs with complete descriptions rather than sparse summaries. Code review becomes faster when the reviewer understands what they are looking at before they read the diff.
How enterprises use this. Build an audit trail of every change with consistent, structured explanations. Engineering governance becomes less dependent on individual engineer discipline.
Feature 16: CLI and IDE integrations
Codex works across VS Code, JetBrains, terminal and CLI, desktop apps, and GitHub workflows. You do not need to change your development environment to use it. It integrates into the workflow you already have.
How engineers use this. Access Codex capabilities without leaving your editor or your terminal. The tool comes to where you work rather than requiring you to go to a separate interface.
Feature 17: Security and vulnerability analysis
Codex Security can detect vulnerabilities, propose fixes, build threat models, and validate security findings across your codebase.
Security analysis is one of the most consistently under-resourced parts of the software development lifecycle. Codex makes it possible to run a security review on every change rather than periodically or on selected releases.
How DevSecOps teams use this. Embed security analysis into the pull request workflow. Every PR is reviewed for vulnerabilities before it merges. The security review that previously happened quarterly or on high-priority releases happens continuously.
How enterprises use this. Meet compliance requirements with automated, documented security reviews. The audit trail is generated automatically as part of the development workflow.
Part Four: Who uses what and how
The product manager
ChatGPT is the PM’s reasoning partner, research engine, and writing collaborator.
Conversational reasoning for product decisions. Deep research for market intelligence. File understanding for synthesising user research. Structured document generation for PRDs, user stories, and stakeholder updates. Persistent memory so every output is grounded in your actual product context.
Codex becomes available to PMs who want to commission working prototypes. Give it a spec. Review the pull request. Validate the approach before committing engineering time. The distance between a product idea and working software shrinks dramatically.
The software engineer
Codex handles the implementation layer. Feature development, bug fixes, refactoring, test generation, documentation, and PR creation. The engineer’s role shifts from writing code to directing the agent and reviewing the output.
ChatGPT handles the thinking layer. Architecture trade-off analysis. Technical design discussions. Debugging complex problems. Reasoning through system behaviour. The work that requires thinking rather than typing.
Together they change what a single engineer can accomplish in a day. Multiple Codex agents running in parallel, each delivering a PR for review. ChatGPT helping reason through the architectural decisions that shape what gets built.
The solutions architect
ChatGPT is the architect’s design partner. System design conversations, trade-off analysis, cloud architecture patterns, integration architecture, and AI system design all benefit from a reasoning partner that understands the full breadth of options.
Multimodal understanding means architects can bring diagrams, screenshots, and whiteboards into the conversation rather than describing them. The depth of analysis improves when the AI can see the system rather than hear about it.
Codex handles the prototyping layer. When an architecture decision needs validation, Codex can implement a working version quickly. The architect’s intuition gets tested against working code rather than staying theoretical.
The technical PM
Technical PMs sit at the intersection of product and engineering. ChatGPT serves both sides of that position simultaneously.
On the product side: research, discovery, PRDs, stakeholder communication. On the engineering side: API design reasoning, system trade-off analysis, technical documentation, architecture evaluation. The tool that previously required switching between a business mindset and a technical one now supports both in the same conversation.
API product management specifically benefits significantly. API lifecycle strategy, versioning governance, developer experience, platform strategy, SDK planning, and integration mapping are all topics where ChatGPT’s reasoning depth is particularly strong.
The startup founder
Small founding teams can now operate at the throughput of much larger organisations.
ChatGPT handles research, strategy, writing, and product thinking. Codex handles engineering acceleration, boilerplate generation, and feature implementation. The operational overhead of building, the documentation, the communication, the research, the analysis, runs largely on AI.
This is not hypothetical. It is the reason AI-native startups are achieving in months what previously took years. The leverage is real and it compounds with every tool and workflow you embed it in.
The enterprise
Enterprise ChatGPT supports shared workspaces, organisational memory, admin governance, and team collaboration. The security and compliance layer includes SOC2 controls, secure environments, permission-aware retrieval, and enterprise data governance.
The connector ecosystem integrates with GitHub, Slack, Google Drive, SharePoint, Notion, and Jira environments. Knowledge retrieval across internal documentation, repositories, issues, and PRs becomes a capability rather than a manual research exercise.
At scale the consistency gains are as valuable as the speed gains. Every team operating with the same tools, the same quality floor, and the same access to reasoning support. The strongest individuals stop being a bottleneck because the capability is distributed.
Part Five: The bigger shift
From assistants to autonomous collaborators
The biggest shift happening right now is not about any individual feature.
ChatGPT and Codex are evolving from assistants you query to autonomous collaborators you direct. From tools that respond to tools that act. From software that generates a suggestion to systems that deliver a completed output.
This changes three things simultaneously.
Software development shifts from manual implementation to orchestration, supervision, architecture, and validation. The engineer’s work becomes less about writing code and more about the decisions that shape what gets built and the review that ensures it is built correctly.
Product management shifts from document management to AI-augmented strategic operation. PMs can prototype faster, generate specs instantly, analyse markets continuously, automate operational work, and simulate strategic decisions. The role becomes more strategic because the execution layer runs faster.
Solutions architecture shifts from slow iteration to rapid prototyping and instant architecture comparison. Decisions that previously required days of documentation and review can be explored and validated in hours.
The role that remains yours
Understanding these tools completely does not mean outsourcing your judgement to them.
The decisions that matter still belong to the people who understand the context. The strategy. The user. The constraints. The values. What is worth building and why.
AI handles the execution layer faster and faster. The people who understand systems well enough to direct that execution effectively are the ones who create the most value in the organisations around them.
That has been the argument of this entire series. The fundamentals are not less important because AI can implement them. They are more important because the quality of your direction determines the quality of what gets built.
Reflection for the week
We have now covered Claude Code, Claude Cowork, ChatGPT, and Codex in full.
Four tools. Dozens of features. One consistent pattern underneath all of them.
The people who understand what they are asking for get dramatically more from these tools than the people who do not. The quality of the context determines the quality of the output. The understanding of the system determines the quality of the direction.
This is what the entire Web 4.0 Basics series has been building toward. Not a list of tools. An understanding of the layer beneath the tools. So that every feature, every capability, and every workflow update that follows lands on a foundation that makes sense.
The question worth sitting with this week:
Looking at your role as it stands today, which one workflow still runs entirely on manual effort and human memory? And which of these features would change that the most?
Want to build this foundation properly?
Everything in this series, from how the internet works to how to direct ChatGPT and Codex effectively, is what we teach in our 14-week Amakora AI Product and Systems Fellowship starting next month.
Live classes. Real projects. A cohort of people building alongside you.
Apply here: amakoragroup.com/apply
Until next week,
Tochii
Founder, Learn with Tochii
Inspire · Educate · Empower
📧 contact@tochukwuachebe.com
🌐 https://www.tochukwuachebe.com