Prasad Thilakarathne (@prasadct) header
Prasad Thilakarathne (@prasadct) profile picture

Software Engineer | NUS MSc (AI & Innovation) | Building real-world AI products

Singapore
Joined June 2011
Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)Photo by Prasad Thilakarathne (@prasadct)
Dec 2021: Log4Shell ruined Christmas for Java devs. Nov 2026: Java4Shell, a 10-year-old OpenJDK bug found with AI. 100,000+ CVEs this year, double 2025. AI made finding bugs cheap. Patching is still manual.
4
Replying to @chamath
@chamath said AI labs are the fridge makers and the real money is in CocaCola. 2026: the fridge makers now sell Coke, Pepsi, the vending machine, and a browser to find more Coke. If you're building on their APIs, check what they ship next week.
1
27
Last 7 days alone: • Anthropic: Claude Meetings, Sheets, Whiteboards • OpenAI: Daybreak AI pentesting • Google: Gemini on CarPlay, Universal Cart, Meeting bots, AI spreadsheets, AI security, in-car voice. All startup categories last quarter.
Most startups die from building the wrong thing, not from bad code. WorkBack AI runs Amazon's Working Backwards method in 23 steps, and workspaces make every solution step judge ideas against what your company can build and won't. workbackai.com
20
My Claude Code prompts aren’t prompts. They’re GitHub issues. Plan the feature, write the issues, let the agent build them one by one, review each PR. Chat history vanishes. Issues stay.
🤖 Made with AI
1
2
41
Built a policy-based AI agent for Java developers on weekends. Slack → reads GitHub issue → fixes the bug → asks for approval → creates the PR. That approval gate is the whole point. The agent asks before it acts. #Java #SpringBoot #SpringAI #Opensource
2
40
Gave Claude Code a PID and said "something's leaking, investigate." It ran jstack, ps, lsof on its own. Found blocked threads, worker pool exhaustion, event loop starvation. Fixed the code. Then added a circuit breaker I never asked for. prasadct.com/blog/claude-cod… #ClaudeCode #AI
1
31
“Simplicity is a prerequisite for reliability.” - Edsger Dijkstra AI can generate 1000 lines in seconds. The skill now is knowing which 50 lines actually matter.
7
Problem-solving is not about finding answers quickly. It's about asking better questions slowly. AI excels at finding answers quickly. But you need to own the responsibility of asking better questions. Don't outsource your thinking just because you can outsource your typing.#AI
1
11
The best software engineers that I know use AI for: 1. Explore edge cases they didn’t have time to consider before 2. Prototype failure modes 3. Stress-test at design time 4. Draft boilerplate #AI #SoftwareEngineering #SeniorSoftwareEngineer #AIAssistedCoding
16
The real skill now is curation, not generation. AI gives quantity, humans give meaning. #AI #Innovation #Creativity
19
GenAI, Is it a tool? A collaborator? A simulator? In my experience working with these systems, the answer is "all of the above," depending on how we choose to engage with it. This ambiguity is both exciting and unsettling. #GenAI #ProblemSolving #Innovation #Software
19
From leading distributed Java systems to ML engineering. Used to hunt bugs affecting 4% of production requests. Now hunting 4% fraud in 11K transactions. Same detective brain, different domain! 🔍 #MachineLearning #MSc #SoftwareEngineering
16
Prasad Thilakarathne retweeted
We need a new term for when you vibe code something but delete all the LLM-generated comments to make it look like you wrote the code yourself 😂
470
116
68
3,807
379,177
Amara’s Law says: we overestimate a technology in the short run and underestimate it in the long run. AI is proving this right. At first, we get excited (and sometimes disappointed) by quick demos. But the real impact will show years later, quietly changing how we live and work.
15
Why 95% of AI projects fail? It’s not because the tech is weak. It’s because of the basics are skipped. No clear problem to solve, No measurable success, No Problem-solving or Design Thinking.Just because it’s AI people expect magic. But AI isn’t magic. It’s all Engineering. #AI
1
14
Listening to the NotebookLM podcast, sometimes it’s hard to accept if it’s AI or a human. The Q&A feature makes it even more interesting, it feels like having a real conversation. #NotebookLM #GenAI #AI #prasadct
21